CN114442079B - Fall detection method and device for target object - Google Patents

Abstract

The invention discloses a method and a device for detecting falling of a target object. Wherein the method comprises the following steps: receiving a feedback signal generated by a transmitting signal of a target object transmitting radar; determining corresponding range-doppler information from the feedback signal; determining human body velocity information of the target object through the distance-Doppler information; in case the human speed information exceeds a preset speed threshold, it is determined whether the target object falls according to the range-doppler information. The invention solves the technical problems of large calculation amount, high cost, high false alarm rate and poor privacy of falling detection in the prior art.

Description

Method and device for detecting fall of target object

Technical Field

The present invention relates to the field of fall detection, and in particular to a fall detection method and device for a target object.

Background Art

As China gradually enters an aging society and young people face high work pressure, the number of empty-nest elderly people is increasing. The fact that the elderly accidentally fall and no one finds them leads to the lack of timely treatment and eventually causes casualties, which is an important reason affecting the physical and mental health of the elderly. In the bathroom environment, due to the slippery floor and small space, physical discomfort such as dizziness after showering or sitting on the toilet is more likely to cause falls. The fall detection solution based on millimeter-wave radar has the advantages of strong penetration in the bathroom environment, not being easily affected by environmental factors such as fog and light, and strong privacy protection. It is the mainstream choice for current applications.

Solutions of existing technology: There are currently four types of solutions for detecting bathroom falls: three-dimensional point cloud imaging using radar equipment, multi-sensor solutions such as infrared + radar, radar equipment for determining the speed and height information of the target, and camera imaging solutions.

Existing problems and defects: 3D point cloud imaging requires more antennas to have better angular resolution, the equipment cost is high, and the point cloud information cannot restore the human body posture well after clustering, and there are still problems of misjudgment and missed judgment. The infrared device will fail when showering, which affects the installation convenience, cost and system robustness of the multi-sensor system. In the complex usage scenarios of the bathroom, fall detection based on speed and height information dimensions can easily cause many misjudgments. The camera imaging solution has the problem of invading people's privacy in private spaces such as the bathroom.

To address the above-mentioned problems, no effective solution has been proposed yet.

Summary of the invention

The embodiments of the present invention provide a method and device for detecting a fall of a target object, so as to at least solve the technical problems of the prior art fall detection, such as large amount of calculation, high cost, high false alarm rate and missed alarm rate, and poor privacy.

According to one aspect of an embodiment of the present invention, a method for detecting a fall of a target object is provided, comprising: receiving a feedback signal generated by a transmission signal of a radar transmitted by the target object; determining corresponding distance-Doppler information according to the feedback signal; determining human body speed information of the target object through the distance-Doppler information; and determining whether the target object has fallen according to the distance-Doppler information when the human body speed information exceeds a preset speed threshold.

Optionally, before receiving the feedback signal generated by the transmission signal of the radar transmitted by the target object, the method also includes: detecting normal posture parameters of the target object in the detection environment and environmental parameters of the detection environment through the radar transmission signal at a preset frequency; matching the normal posture parameters with credible data in a posture database, wherein the posture database stores credible normal posture parameters and environmental parameters; writing the normal posture parameters into the posture database if the match is successful; and updating the posture threshold parameters according to the latest data in the posture database, wherein the posture threshold parameters are used to detect whether the target object has fallen.

Optionally, before receiving the feedback signal generated by the transmission signal of the radar transmitting the target object, the method also includes: detecting the ambient noise power when the target object is not in the detection environment through the transmission signal of the radar, wherein the posture threshold parameter includes the ambient noise power; determining the signal average power of the feedback signal on the floor of the detection environment through the feedback signal; determining whether the target object exists in the detection environment based on the ambient noise power and the signal average power; wherein, when the signal average power is greater than the ambient noise power, it is determined that the target object exists in the detection environment.

Optionally, determining the corresponding distance-Doppler information according to the feedback signal includes: performing a fast Fourier transform on a fast-time signal of the feedback signal after removing DC to obtain distance dimension information; performing a fast Fourier transform on a slow-time signal of the feedback signal after removing DC to obtain Doppler information; and accumulating the distance dimension information according to the distance dimension information and the Doppler information to obtain the distance-Doppler information.

Optionally, the fast time signal of the feedback signal is subjected to a fast Fourier transform after DC removal to obtain distance dimension information, including: performing a fast Fourier transform on the fast time signal of each frame signal of the feedback signal in multiple slow time dimensions to obtain a first distance image; performing a fast Fourier transform on the fast time signal of the feedback signal in the frame time dimension to obtain a second distance image; wherein the distance dimension information includes the first distance image and the second distance image.

Optionally, when the human body speed information exceeds a preset speed threshold, determining whether the target object falls according to the distance-Doppler information includes: detecting the height of the target object according to the distance-Doppler information, and determining a first risk probability of the target object falling; detecting the distance gate of the floor of the detection environment where the target object is located according to the distance-Doppler information, and whether there is a human breathing frequency characteristic, and determining a second risk probability of the target object falling; determining envelope graphs of different distance gates according to the distance-Doppler information, and determining a third risk probability of the target object falling; determining a fourth risk probability of the target object falling according to whether the distance-Doppler information has a weak breathing signal characteristic; determining a comprehensive probability of the target object falling according to the first risk probability, the second risk probability, the third risk probability and the fourth risk probability, and their corresponding weights; when the comprehensive probability reaches a preset probability threshold, determining that the target object has fallen.

Optionally, detecting the height of the target object according to the range-Doppler information, and determining a first risk probability of the target object falling includes: determining whether the target object has movement close to the ground according to the first range image; in the case of having the movement close to the ground, obtaining a corresponding fall height according to the human body height in the posture threshold parameter, wherein the fall height corresponds to the human body height; the human body height is determined based on the position of the high power point in the second range image at the range gate and the installation height of the radar; determining the highest height of the target object according to the first range image; and in the case where the highest height is less than the fall height, determining that the target object has the first risk probability of falling.

Optionally, the distance gate of the ground of the detection environment where the target object is located is detected according to the distance-Doppler information to see whether there are human breathing frequency characteristics, and determining the second risk probability of the target object falling includes: performing a fast Fourier transform on the distance gate within a first preset height range of the floor according to the second distance image to obtain first target frequency information; when the first target frequency information has a breathing frequency characteristic, determining that the target object has the second risk probability of falling.

Optionally, determining envelope graphs of different distance gates according to the distance-Doppler information, and determining a third risk probability of the target object falling includes: determining envelope graphs of different distance gates according to the first distance image; and when the envelope graph does not match the envelope graph of the target object when not falling, determining that the target object has the third risk probability of falling, wherein the posture threshold parameter includes the envelope graph of the normal posture of the target object when not falling.

Optionally, determining the fourth risk probability of the target object falling according to whether the distance-Doppler information has a signal characteristic of weak breathing includes: performing a fast Fourier transform on a range gate within a second preset height range according to the second distance image to obtain second target frequency information, wherein the second preset height range is higher than the first preset height range; in case the second target frequency information has a breathing frequency characteristic, determining that the target object has the fourth risk probability of falling.

According to another aspect of an embodiment of the present invention, a fall detection device for a target object is also provided, including: a receiving module, used to receive a feedback signal generated by a transmission signal of a radar transmitted by the target object; a first determination module, used to determine corresponding distance-Doppler information according to the feedback signal; a second determination module, used to determine human body speed information of the target object through the distance-Doppler information; and a third determination module, used to determine whether the target object falls according to the distance-Doppler information when the human body speed information exceeds a preset speed threshold.

According to another aspect of an embodiment of the present invention, a processor is further provided, and the processor is used to run a program, wherein the program executes any one of the above-mentioned methods for detecting a fall of a target object when running.

According to another aspect of an embodiment of the present invention, a computer storage medium is further provided, wherein the computer storage medium includes a stored program, wherein when the program is executed, the device where the computer storage medium is located is controlled to execute any one of the above-mentioned methods for detecting a fall of a target object.

In an embodiment of the present invention, a feedback signal generated by a transmission signal of a radar transmitted by a target object is received; corresponding distance-Doppler information is determined according to the feedback signal; human body speed information of the target object is determined through the distance-Doppler information; when the human body speed information exceeds a preset speed threshold, whether the target object falls is determined according to the distance-Doppler information. The distance Doppler information is obtained through the feedback signal of the radar, and when the human body speed information exceeds the preset speed threshold, it is determined whether the target object falls, thereby achieving the purpose of detecting whether a fall is performed by obtaining distance-Doppler information through radar monitoring, not only avoiding the problem of large amount of calculation in radar detection through a three-dimensional point cloud method, avoiding the high cost of infrared detection devices, and high false alarm rate and missed alarm rate when environmental conditions are harsh, but also avoiding the problem of poor privacy of camera detection, achieving the technical effects of reducing the amount of calculation, reducing the cost, improving the accuracy and improving the privacy, thereby solving the technical problems of large amount of calculation, high cost, high false alarm rate and missed alarm rate, and poor privacy in the prior art of fall detection.

BRIEF DESCRIPTION OF THE DRAWINGS

The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

FIG1 is a flow chart of a method for detecting a fall of a target object according to an embodiment of the present invention;

FIG2 is a schematic diagram of a detection system architecture according to an embodiment of the present invention;

FIG3 is a flow chart of an overall detection method according to an embodiment of the present invention;

FIG4 is a flow chart of data iteration of a SaaS system according to an embodiment of the present invention;

FIG5 is a flow chart of a fall detection algorithm according to an embodiment of the present invention;

FIG. 6 is a schematic diagram of a fall detection device for a target object according to an embodiment of the present invention.

DETAILED DESCRIPTION

In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

According to an embodiment of the present invention, a method embodiment of a fall detection method for a target object is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

FIG. 1 is a flow chart of a method for detecting a fall of a target object according to an embodiment of the present invention. As shown in FIG. 1 , the method includes the following steps:

Step S102, receiving a feedback signal generated by a transmission signal of a radar transmitted by a target object;

Step S104, determining corresponding range-Doppler information according to the feedback signal;

Step S106, determining the human body velocity information of the target object through range-Doppler information;

Step S108: When the human body speed information exceeds a preset speed threshold, determining whether the target object falls according to the distance-Doppler information.

Through the above steps, a feedback signal generated by a transmission signal of a radar transmitted by a target object is received; corresponding distance-Doppler information is determined according to the feedback signal; human body speed information of the target object is determined according to the distance-Doppler information; when the human body speed information exceeds a preset speed threshold, whether the target object falls is determined according to the distance-Doppler information. The distance Doppler information is obtained through the feedback signal of the radar, and when the human body speed information exceeds the preset speed threshold, it is determined whether the target object falls, thereby achieving the purpose of detecting whether a fall is obtained by obtaining distance-Doppler information through radar monitoring, not only avoiding the problem of large amount of calculation in radar detection through a three-dimensional point cloud method, avoiding the high cost of infrared detection devices, and high false alarm rate and missed alarm rate under harsh environmental conditions, but also avoiding the problem of poor privacy of camera detection, achieving the technical effects of reducing the amount of calculation, reducing the cost, improving the accuracy and improving the privacy, thereby solving the technical problems of large amount of calculation, high cost, high false alarm rate and missed alarm rate, and poor privacy in the prior art of fall detection.

The above radar can be a millimeter wave radar, which sends a detection signal to the target object through a transmitting antenna, for example, ultrasonic waves. After being reflected by the target object, the reflected feedback signal can be received by the receiving antenna of the radar. The fast time signal in the feedback signal can be converted into a one-dimensional distance image signal, that is, the above distance dimension information, through fast Fourier transform. The one-dimensional distance image is the vector sum of the target scattering point sub-echoes projected on the radar ray obtained by the broadband radar according to the signal. It is actually a scattering intensity distribution diagram of each distance unit on the target object. The distance dimension information can obtain the height, speed, etc. of the target object. Specifically, the distance between the target object and the radar can be determined according to the distance dimension information, thereby determining the height of the target object. The height change of the target image over time can be determined through the distance dimension information of multiple frames of images, and then the speed change of the target object can be determined.

In addition, the Doppler dimension information can be obtained by fast Fourier transforming the slow time signal of the feedback signal. Doppler dimension information can detect the weak movement behavior of the target object, including breathing and heartbeat. On the other hand, the distance dimension information and Doppler information contain some overall features. For example, a high-power distance gate can fit an envelope graph that matches the human body posture.

The human body speed information of the target object is determined by the distance-Doppler information. When the human body speed information exceeds the preset speed threshold, it means that the target object has undergone a significant height change movement, and there is a high possibility of falling. Then, other posture parameters are determined based on the distance-Doppler information, including the above-mentioned breathing rate, human body height, and envelope graph, to determine whether the target object has fallen. The specific detection method corresponding to each posture parameter is different, and the target object is comprehensively judged from multiple angles of the above-mentioned multiple posture parameters to determine whether it has fallen. Not only can the accuracy of fall detection be directly improved, but also various false alarms can be avoided from multiple aspects.

This embodiment can measure the radial relative motion speed of the target to the radar according to the Doppler frequency. Falling is a motion away from the radar, and its Doppler frequency presents a negative value. After selecting a suitable threshold, if it exceeds the threshold, it is considered to be a falling motion.

The distance Doppler information is obtained through the feedback signal of the radar. When the human body speed information exceeds the preset speed threshold, it is determined whether the target object has fallen according to the distance Doppler information, thereby achieving the purpose of detecting whether a fall has been performed by obtaining distance-Doppler information through radar monitoring. This not only avoids the problem of large amount of calculation in radar detection through three-dimensional point cloud, the high cost of infrared detection devices, high false alarm rate and missed alarm rate under harsh environmental conditions, but also avoids the problem of poor privacy in camera detection, and achieves the technical effect of reducing the amount of calculation, reducing the cost, improving the accuracy and improving the privacy, thereby solving the technical problems of large amount of calculation, high cost, high false alarm rate and missed alarm rate, and poor privacy in the prior art of fall detection.

Optionally, before receiving the feedback signal generated by the transmission signal of the radar transmitted by the target object, the method also includes: detecting normal posture parameters of the target object in the detection environment and environmental parameters of the detection environment through the radar transmission signal at a preset frequency; matching the normal posture parameters with trusted data in a posture database, wherein the posture database stores trusted normal posture parameters and environmental parameters; writing the normal posture parameters into the posture database if the match is successful; and updating the posture threshold parameters according to the latest data in the posture database, wherein the posture threshold parameters are used to detect whether the target object has fallen.

The above-mentioned posture threshold parameters include the ambient noise power when the subsequent target object is not in the detection environment, and also include the height of the target object, the envelope diagram of the target object in a normal posture without falling, including standing posture, sitting posture, etc. The above-mentioned posture threshold parameters can provide parameter basis for subsequent detection, and the above-mentioned posture threshold parameters are required for risk determination of multiple parameters in subsequent fall detection. The normal posture parameters of the target object can be determined by normal posture parameters and environmental parameters, and then the posture parameters under normal conditions can be processed to obtain other threshold parameters. For example, by processing normal distance-Doppler information, the envelope diagram of normal posture, the height range of human body in normal posture, etc. are obtained.

It should be noted that the posture threshold parameters of the posture database can also be updated through specific fall detection. After a fall detection, after the user confirms, the trusted data in the detection process is written into the posture database, and the posture threshold parameters are updated. That is, in other embodiments, the posture threshold parameters can also include posture parameters of falling, so that the posture parameters of falling can be matched with the posture parameters of the detected target object from the perspective of the posture parameters of falling to determine the degree of proximity. If they are close enough, it can also be determined that the target object has fallen.

In this embodiment, as shown in FIG4 , the above-mentioned posture database is a posture database of a SaaS system. The SaaS system adapts the human posture database information and accumulates and iterates the posture threshold parameters in the fall detection and recognition algorithm according to the data used by daily users. Step S31, determine whether the data accumulation is full for three days. If not, load the default parameters. If satisfied, enter step S32. Step S32, extract the credible data in the database through information such as mean and variance, and compare it with the current storage result. Step S33, if the current data has a low degree of match with the historical data in the database, abandon the information storage of this calculation. If the confidence is high, write it into the database. Step S34, load the parameters such as radar installation height, environmental noise power, user height information, and human low posture information, and provide them for subsequent algorithm modules. The above-mentioned posture database is updated in a timely manner, so that the posture threshold parameters in the posture database are updated in a timely manner according to the situation, thereby improving the accuracy of the target object falling.

Optionally, before receiving the feedback signal generated by the transmission signal of the radar transmitted by the target object, the method also includes: detecting the ambient noise power when the target object is not in the detection environment through the transmission signal of the radar, wherein the posture threshold parameter includes the ambient noise power; determining the average signal power of the feedback signal on the floor through the feedback signal of the floor of the detection environment; determining whether there is a target object in the detection environment based on the ambient noise power and the signal average power; wherein, when the signal average power is greater than the ambient noise power, it is determined that there is a target object in the detection environment.

Before receiving the feedback signal generated by the transmission signal of the radar transmitting the target object, it is also possible to determine whether there is a target object in the detection environment through the transmission signal of the radar. Specifically, the average signal power of the floor is compared with the ambient noise power. If the average signal power is greater than the ambient noise power, it is determined that there is a target object in the detection environment. Thus, when it is determined that the target object has entered the detection environment, the feedback signal of the above radar is obtained.

In this embodiment, the average power of the Micro distance image of the feedback signal is calculated near the floor, and if the difference between the average power and the noise power exceeds the set threshold range, it is considered that there is a person. Therefore, when it is determined that there is a person, according to the above steps, the purpose of detecting whether a person has fallen is achieved by obtaining distance-Doppler information through radar monitoring, which avoids the problem of poor accuracy in performing the detection according to the above steps at regular intervals, and also avoids the problem of executing the above steps in the absence of people, which will lead to a waste of computing resources.

Optionally, determining the corresponding distance-Doppler information based on the feedback signal includes: performing a fast Fourier transform on a fast-time signal of the feedback signal after removing the DC to obtain distance dimension information; performing a fast Fourier transform on a slow-time signal of the feedback signal after removing the DC to obtain Doppler information; and accumulating the distance dimension information based on the distance dimension information and the Doppler information to obtain the distance-Doppler information.

Specifically, read the original data of the radar echo signal from the radar's ADC buffer data. Perform FFT on the fast-time signal after removing the DC component to obtain the distance dimension information. Perform FFT on the slow-time signal after removing the DC component to obtain the Doppler dimension information. For each frame of the Micro range image, accumulate 20 frames of data and remove the DC component to obtain the average value to obtain the Micro range image. The range Doppler image is obtained by performing another FFT on the Micro. The Micro one-dimensional range image only contains distance and energy information. The range Doppler contains distance, speed, and energy information.

For the above-mentioned fast time signal and slow time signal, the radar sends pulse signals periodically when working, and samples the echo signal within the pulse interval. Although the echo sampling interval and the pulse repetition interval (pulse period) are on the same time axis, they are very different in magnitude. For example, the echo sampling interval is about 10-8, while the pulse repetition interval is about 10-3. Therefore, the echo sampling interval and the pulse repetition interval are divided into two dimensions, which are called fast time and slow time, and their corresponding signals are also fast time signals and slow time signals.

Optionally, the fast time signal of the feedback signal is subjected to a fast Fourier transform after DC removal to obtain distance dimension information, including: performing a fast Fourier transform on the fast time signal of each frame signal of the feedback signal in multiple slow time dimensions to obtain a first distance image; performing a fast Fourier transform on the fast time signal of the feedback signal in the frame time dimension to obtain a second distance image; wherein the distance dimension information includes the first distance image and the second distance image.

The above-mentioned distance dimension information includes the above-mentioned first distance image and the second distance image, and the above-mentioned feedback signal may include multiple frame signal segments, and the time corresponding to each frame signal segment may be the same. Each frame signal of the above-mentioned feedback signal performs a fast Fourier transform on the fast time signal in multiple slow time dimensions to obtain a first distance image, that is, a Macro distance image, which is a fast Fourier transform of the fast time signal in multiple slow time dimensions of each frame. It is actually a fast Fourier transform in the signal segment of each frame, which can detect distance changes with small movement amplitudes, such as body ups and downs caused by breathing. The above-mentioned fast Fourier transform of the fast time signal of the feedback signal in the frame time dimension to obtain the second distance image, that is, the Micro distance image, can be determined separately The fast time signal of each frame of the feedback signal is subjected to a fast Fourier transform, and then combined in order to obtain the second distance image. Since the second distance image is in the frame time dimension, it can detect distance changes with large movement amplitudes, such as changes in human height caused by human movement.

Optionally, when human body speed information exceeds a preset speed threshold, determining whether the target object falls according to the distance-Doppler information includes: detecting the height of the target object according to the distance-Doppler information, and determining a first risk probability of the target object falling; detecting the distance gate of the floor of the detection environment where the target object is located according to the distance-Doppler information to see whether there is a human breathing frequency characteristic, and determining a second risk probability of the target object falling; determining envelope graphics of different distance gates according to the distance-Doppler information, and determining a third risk probability of the target object falling; determining a fourth risk probability of the target object falling according to whether the distance-Doppler information has a weak breathing signal characteristic; determining a comprehensive probability of the target object falling according to the first risk probability, the second risk probability, the third risk probability and the fourth risk probability, and their corresponding weights; and determining that the target object falls when the comprehensive probability reaches a preset probability threshold.

The distance dimension information of the above distance-Doppler information extracts human height information, adaptively matches the corresponding fall height for different body heights, and improves the applicability of the algorithm. The Micro second distance image extracts human breathing information near the floor, that is, the above human breathing frequency characteristics, to eliminate interference and misjudgment caused by complex environments in the bathroom, such as toilets and water pipes. The Micro second distance image extracts human envelope information in the distance dimension, that is, the above envelope graph, which can effectively eliminate misjudgment in scenes with lower postures such as sitting on the toilet. The Macro first distance image extracts human activity information in the distance dimension, including the above weak breathing signal characteristics, which can eliminate misjudgment in scenes such as hand washing clothes in the bathroom.

Based on the above risk judgment, the probability of human falling is weighted and calculated. If a signal that is obviously not a fall is found in the process of accumulating the fall probability, such as a signal at a higher distance door, the person must be standing, and the previously accumulated fall probability is a misjudgment, so it is cleared and corrected. If the accumulated fall probability exceeds the threshold, a fall alarm message is output.

Optionally, detecting the height of the target object according to the distance-Doppler information and determining the first risk probability of the target object falling includes: determining whether the target object has movement close to the ground according to the first distance image; in the case of movement close to the ground, obtaining the corresponding fall height according to the human body height in the posture threshold parameter, wherein the fall height corresponds to the human body height, which is also the human body height of the target object; the human body height is determined according to the position of the high-power point in the second distance image at the range gate and the installation height of the radar; determining the highest height of the target object according to the distance dimension information; and in the case where the highest height is less than the falling height descent distance range, determining that the target object has a first risk probability of falling.

The distance-Doppler image of the Macro distance image (also known as the first distance image) is used to extract whether the human body has a movement away from the radar, that is, falling to the ground. The Doppler frequency set by the threshold should not be too high to avoid underreporting of slow fall scenes due to dizziness. After detecting the trend of the human body moving away from the radar, the Micro distance image (also known as the second distance image) can be used to calculate the human body height information through the position of the high power point in the one-dimensional distance image at the distance gate and the radar installation height. According to the human body height information, the corresponding human body falling height descent distance summarized by big data is matched. If it is found that the non-noise power distance gate closest to the radar is far from the normal range of the human body descent height, it is considered that this person has the first risk probability of falling. Adaptively match the corresponding falling height for different heights to improve the applicability of the algorithm.

Optionally, the distance gate on the ground of the detection environment where the target object is located is detected according to the distance-Doppler information to see whether there is a human respiratory frequency characteristic, and determining the second risk probability of the target object falling includes: performing a fast Fourier transform on the distance gate within a preset height range of the floor according to the Micro range image (second range image) to obtain target frequency information; when the target frequency information has a clear respiratory frequency characteristic, determining that the target object has a second risk probability of falling.

After a person falls, he or she will lie on the floor. When the radar height is known, the FFT of the distance door sliding window near the floor is performed to extract the frequency information of the signal. When a person is doing other normal activities in the bathroom, the frequency domain is relatively messy. When a person falls, the frequency characteristics of a breath can be extracted relatively clearly, and it is considered that this person has a higher probability of falling again. This can effectively eliminate interference misjudgment caused by complex environments in the bathroom, such as toilets and water pipes.

Optionally, determining envelope graphics of different distance gates according to the distance-Doppler information, and determining the third risk probability of the target object falling includes: determining envelope graphics of different distance gates according to the first distance image; when the envelope graphics do not match the envelope graphics of the target object not falling, determining that the target object has a third risk probability of falling, wherein the posture threshold parameter includes the envelope graphics of the normal posture of the target object without falling.

The standing and sitting postures of a person will present an envelope graph that spans multiple range gates and matches the posture of the person on the one-dimensional range image of the Macro first range image. When a person falls, a very narrow envelope graph with only a few range gates having strong energy will appear on the one-dimensional range image. When such an envelope graph feature appears, it is considered that the person has a third risk probability of falling. This can effectively eliminate misjudgments in scenes with low postures such as sitting on the toilet.

Optionally, determining a fourth risk probability of the target object falling according to whether the distance-Doppler information has a signal characteristic of weak breathing includes: performing a fast Fourier transform on a range gate within a second preset height range according to the second distance image to obtain second target frequency information, wherein the second preset height range is higher than the first preset height range; and determining that the target object has the fourth risk probability of falling when the second target frequency information has a breathing frequency characteristic.

In the scenario where a person cannot save himself after fainting, a person who falls on a toilet or sink can see very obvious signals due to weak breathing movements in the Micro one-dimensional distance image (second distance image), but almost no signals can be seen in the Macro one-dimensional distance image (first distance image) due to the absence of rapid movements. This means that the person has a high fourth risk probability of falling. This can eliminate misjudgments in scenarios such as hand washing in the bathroom.

It should be noted that this embodiment also provides an optional implementation, which is described in detail below.

This embodiment proposes a bathroom fall detection method based on radar one-dimensional range image to solve the problems of large computational complexity, high equipment cost, high false alarm rate and missed alarm rate, privacy infringement and susceptibility to environmental influences in existing fall detection methods.

FIG3 is a flow chart of an overall detection method according to an embodiment of the present invention. As shown in FIG3 , this embodiment includes the following steps:

Step S1: radar is deployed in the bathroom. The radar is required to be installed in the middle of the bathroom ceiling, irradiate downward, and conduct air sampling of the environment. The distance information of the radar from the ground, signal noise information and other parameters are calculated and uploaded to the cloud.

Step S2, parameter initialization, configure the FMCW signal starting frequency, cutoff frequency, rising edge frequency modulation slope, duration, falling edge duration, rest time; configure the ADC sampling frequency, fast time sampling points, chirp number in one frame, frame period;

Step S3: The network provides a software service SaaS system to adapt the human posture database information, and accumulate and iterate the posture threshold parameters in the fall detection and recognition algorithm based on the daily user usage data.

Step S4, reading the radar echo signal raw data from the ADC buffer data.

Step S5, performing FFT on the fast time signal after removing the DC component to obtain distance dimension information.

Step S6, performing FFT on the slow-time signal after removing DC to obtain Doppler dimension information.

Step S7, perform FFT on the fast time in the frame and accumulate and smooth the DC to obtain a Micro one-dimensional range image, which is the second range image mentioned above. Perform FFT on the fast time between frames and accumulate and smooth the DC to obtain a Macro one-dimensional range image, which is the first range image mentioned above.

Step S8, calculate the average power of the Micro distance image near the floor and the door, compare it with the noise power and filter it to get the current toilet presence or absence status. If the difference between the average power of the Micro distance image near the floor and the door exceeds the set threshold range, it is considered to be a occupied state.

Step S9, determine whether there is anyone in the current bathroom, and if no one is there, return to step S4. If there is someone, enter the fall detection information extraction algorithm.

Step S10, upload the user feature information extracted in S9, such as height, toilet height, washing height, etc., to the cloud, and return to step S3.

FIG. 4 is a flowchart of SaaS system data iteration according to an embodiment of the present invention. As shown in FIG. 4 , the SaaS system posture library information of this embodiment includes the following steps:

Step S31, determine whether the data accumulation is three days or more, if not, load the default parameters, if satisfied, proceed to step S32.

Step S32, extracting credible data from the database through information such as mean and variance, and comparing it with the current storage results.

Step S33: If the current data has a low matching degree with the historical data in the database, the information storage of this calculation is abandoned. If the confidence is high, it is written into the database.

Step S34, loading parameters such as radar installation height, environmental noise power, user height information, human low posture information, etc., for use by subsequent algorithm modules.

FIG5 is a flow chart of a fall detection algorithm according to an embodiment of the present invention. As shown in FIG5 , the fall detection information extraction algorithm of this embodiment includes the following steps:

Step S71, Macro distance-Doppler image extracts human body speed information, and enters step S72 after detecting that the speed exceeds the radar speed threshold, otherwise the fall detection information extraction algorithm ends.

Step S72, the Micro distance dimension extracts the height information of the human body, and adaptively matches the corresponding fall height for different heights to improve the applicability of the algorithm. The Micro distance-Doppler dimension extracts the human body breathing information near the floor to eliminate the interference misjudgment caused by the complex environment in the bathroom, such as toilets, water pipes, etc. The Micro distance dimension extracts the human body envelope information, which can effectively eliminate the misjudgment of scenes with lower postures such as sitting on the toilet. The Macro distance dimension extracts the human body activity information, which can eliminate the misjudgment of scenes such as hand washing clothes in the bathroom.

Step S73, integrate all the information of step S72, weight the information results of different dimensions and calculate the fall score, and accumulate the score of each frame of data.

Step S74, extracting the human standing up information in the Micro distance dimension. If the person has stood up, the fall score counted in step S73 is cleared, otherwise, proceed to step S75.

Step S75, determining whether the score exceeds a threshold value, if so, outputting a fall alarm message, otherwise terminating the fall detection information extraction algorithm.

In the complex environment of the bathroom, this implementation performs multi-dimensional human fall feature information extraction on the one-dimensional distance image of the millimeter-wave radar. In actual use, scenarios that are prone to false alarms, such as sitting on the toilet and bending over to throw things, can effectively avoid false alarms through breathing characteristics, envelope characteristics, etc.

The speed, distance and other information calculated by the radar is highly accurate, and acceleration is not used as the main condition for falling. It has better adaptability to slow falls caused by dizziness in the elderly, and effectively improves the accuracy of fall recognition.

This implementation does not require multiple receiving antennas for angle measurement and imaging, which reduces development costs. The radar is less affected by environmental factors (such as light and fog) and does not infringe on human privacy, making it very suitable for the private environment of the bathroom. There is no need for multiple sensor data to cross-calculate, and the system integration is higher.

During implementation, Figure 2 is a schematic diagram of the detection system architecture according to an embodiment of the present invention. As shown in Figure 2, this embodiment is a bathroom fall detection solution based on millimeter-wave radar, in which the millimeter-wave radar module sends and collects human echo signals, and the processor converts the receiving channel echo signals into digital signals through the A/D chip, and then obtains the fall detection results through the fall detection algorithm module, and outputs the results to the terminal for display through the data transmission module.

Theoretical analysis of implementation methods:

As shown in Figure 3, this millimeter-wave radar fall detection system is installed in the middle of the bathroom ceiling. The millimeter-wave radar is an active sensor. The sensor's transmitting antenna emits pulsed electromagnetic waves at a fixed period. When a person enters the bathroom, the electromagnetic waves are continuously reflected and propagated to the receiving antenna of the millimeter-wave radar sensor. After the sensor obtains the difference frequency signal through mixing, the radar RF signal input by the receiving antenna is filtered and sampled, and then output to the MCU for signal processing.

The MCU first performs a fast-time FFT (Fast Fourier Transform) on the input radar signal in the frame time dimension to obtain the information of the radar signal in the distance dimension, which is called the Macro one-dimensional distance image, and then performs a slow-time FFT to obtain the Doppler dimension information of the radar signal. Then, the fast time is FFTed for multiple slow time dimensions of each frame to obtain the Micro one-dimensional distance image. The Macro dimension measures the rapid movement of the human body, such as the characteristics of a human fall; the Micro dimension can measure the slow movement of the human body. By performing FFT on a specific distance unit of the Micro dimension, micro-speed Doppler information can be obtained, such as the characteristics of human breathing.

By judging the ambient noise and the power difference at the current moment, it can determine whether there is someone in the bathroom. When it is found that there is someone in the bathroom, it enters the detection and recognition algorithm.

As shown in Figure 4, the algorithm uses the distance-Doppler image of the Micro latitude to extract whether the human body has a movement away from the radar, that is, falling to the ground. The Doppler frequency set by the threshold should not be too high to avoid underreporting of slow falling scenes due to dizziness. Specifically, according to the size of the Doppler frequency, the radial relative movement speed of the target to the radar can be measured. Falling is a movement away from the radar, and its Doppler frequency presents a negative value. After selecting a suitable threshold, it is considered that there is a falling movement when it exceeds the threshold. After detecting the trend of the human body moving away from the radar, the human body height information can be inferred from the position of the high-power point in the one-dimensional distance image at the range gate and the radar installation height in the Micro dimension. According to the human body height information, the corresponding human body falling height descent distance under this height is matched after the big data summary. If it is found that the non-noise power range gate closest to the radar is far from the normal range of the human body descent height, it is considered that this person has a risk of falling.

After a person falls, he or she will lie on the floor. When the radar height is known, an FFT is performed on the distance door sliding window near the floor to extract the frequency information of the signal. When a person is doing other normal activities in the bathroom, the frequency domain is relatively messy. After a person falls, the frequency characteristics of a breath can be extracted relatively clearly, and it is considered that this person has a higher risk of falling.

The standing and sitting postures of a person will appear as an envelope graph that spans multiple range gates and matches the posture of the person on the Micro one-dimensional range image. When a person falls, a very narrow envelope graph will appear on the one-dimensional range image, with only a few range gates having strong energy. When such an envelope graph feature appears, it is considered that the person is at risk of falling.

In the scenario where a person cannot save himself after fainting, a very obvious signal caused by the weak movement of breathing can be seen in the Micro one-dimensional distance image after the person falls, but in the Macro dimension, almost no signal can be seen due to the absence of rapid movement. It is considered that this person has a high risk of falling. Micro is sensitive to weak motion detection. The weak movement caused by breathing will reflect power information on the Micro one-dimensional distance image. The power information after falling is mainly concentrated near the floor. Macro is sensitive to rapid motion detection. It can reflect power or Doppler at the moment of falling, but it cannot reflect power after falling. The weak movement of breathing, when falling, the energy is concentrated near the floor (the radar installation height information can be obtained during installation). In other usage scenarios of the bathroom, only the toilet will make the human body relatively still and reflect weak breathing movement, and at this time the energy is concentrated at least 1 meter above the ground. Weak breathing movement will not appear in other usage scenarios.

Based on the above four risk judgments, the probability of a person falling is weighted and calculated. If a signal that is obviously not a fall is found during the accumulation of the fall probability, such as a signal at a higher distance door, the person must be standing, and the previously accumulated fall probability is a misjudgment, so it is cleared and corrected. If the accumulated fall probability exceeds the threshold, a fall alarm message is output.

Every time a user uses the toilet, the user posture data and environmental data measured by the radar will be uploaded to the SaaS system for iterative updates to better learn the environment and human characteristics of the device and improve the performance of fall detection. The database information includes the radar installation height, environmental noise power, user height information calculated by the radar, and low posture height information of daily behaviors such as sitting on the toilet.

In this implementation, the overall architecture of the millimeter-wave radar fall detection algorithm and the fall detection information extraction scheme are key points. In addition, the millimeter-wave radar fall detection data processing process, the implementation steps of the fall detection algorithm, the engineering implementation of the fall detection algorithm, and the detection scheme for weak motion in the Micro dimension, including determining the height of the human body, the energy envelope, and the extraction of breathing characteristics, etc.

FIG6 is a schematic diagram of a fall detection device for a target object according to an embodiment of the present invention. As shown in FIG6 , according to another aspect of an embodiment of the present invention, a fall detection device for a target object is further provided, including: a receiving module 62, a first determination module 64, a second determination module 66 and a third determination module 68. The device is described in detail below.

The receiving module 62 is used to receive the feedback signal generated by the transmitting signal of the radar transmitted by the target object; the first determining module 64 is connected to the above-mentioned receiving module 62, and is used to determine the corresponding distance-Doppler information according to the feedback signal; the second determining module 66 is connected to the above-mentioned first determining module 64, and is used to determine the human body speed information of the target object through the distance-Doppler information; the third determining module 68 is connected to the above-mentioned second determining module 66, and is used to determine whether the target object falls according to the distance-Doppler information when the human body speed information exceeds a preset speed threshold.

Through the above device, a receiving module 62 is used to receive the feedback signal generated by the transmission signal of the radar transmitted by the target object; the first determination module 64 determines the corresponding distance-Doppler information according to the feedback signal; the second determination module 66 determines the human body speed information of the target object through the distance-Doppler information; the third determination module 68 determines whether the target object falls according to the distance-Doppler information when the human body speed information exceeds the preset speed threshold, and obtains the distance Doppler information through the feedback signal of the radar. When the human body speed information exceeds the preset speed threshold, it determines whether the target object falls, thereby achieving the purpose of detecting whether the fall is obtained by obtaining the distance-Doppler information through radar monitoring, not only avoiding the problem of large amount of calculation in radar detection through three-dimensional point cloud, avoiding the high cost of infrared detection device, high false alarm rate and missed alarm rate under harsh environmental conditions, but also avoiding the problem of poor privacy of camera detection, achieving the technical effect of reducing the amount of calculation, reducing the cost, improving the accuracy and improving the privacy, thereby solving the technical problems of large amount of calculation, high cost, high false alarm rate and missed alarm rate, and poor privacy in the prior art fall detection.

According to another aspect of an embodiment of the present invention, a processor is further provided, and the processor is used to run a program, wherein when the program is run, any one of the above-mentioned methods for detecting a fall of a target object is executed.

According to another aspect of an embodiment of the present invention, a computer storage medium is further provided, the computer storage medium including a stored program, wherein when the program is running, the device where the computer storage medium is located is controlled to execute any one of the above-mentioned methods for detecting a fall of a target object.

The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims (12)

1. A method for detecting a fall of a target object, comprising: receiving a feedback signal generated by a target object reflecting a transmission signal of the radar; Determine corresponding range-Doppler information according to the feedback signal; Determine the human body velocity information of the target object through the range-Doppler information; In the case where the human body speed information exceeds a preset speed threshold, determining whether the target object falls according to the distance-Doppler information; When the human body speed information exceeds a preset speed threshold, determining whether the target object falls according to the distance-Doppler information includes: Detecting the height of the target object according to the range-Doppler information, and determining a first risk probability of the target object falling; Detecting the distance gate of the floor of the detection environment where the target object is located according to the distance-Doppler information to determine whether there is a human breathing frequency feature, and determining a second risk probability of the target object falling; Determine envelope graphs of different range gates according to the range-Doppler information, and determine a third risk probability of the target object falling; determining a fourth risk probability of the target object falling according to whether the distance-Doppler information has a signal feature of weak breathing; Determining a comprehensive probability of the target object falling according to the first risk probability, the second risk probability, the third risk probability and the fourth risk probability, and their corresponding weights; When the comprehensive probability reaches a preset probability threshold, it is determined that the target object has fallen. 2. The method according to claim 1, characterized in that before receiving the feedback signal generated by the target object reflecting the transmission signal of the radar, the method further comprises: Detecting normal posture parameters of the target object in the detection environment and environmental parameters of the detection environment by transmitting radar signals at a preset frequency; Matching the normal posture parameters with credible data in a posture database, wherein the posture database stores credible normal posture parameters and environmental parameters; If the matching is successful, writing the normal posture parameters into the posture database; The posture threshold parameter is updated according to the latest data in the posture database, wherein the posture threshold parameter is used to detect whether the target object falls. 3. The method according to claim 2, characterized in that before receiving the feedback signal generated by the target object reflecting the transmission signal of the radar, the method further comprises: Detecting the ambient noise power when the target object is not in the detection environment through the transmission signal of the radar, wherein the attitude threshold parameter includes the ambient noise power; Determine the average signal power of the feedback signal on the floor through the feedback signal of the floor of the detection environment; Determining whether the target object exists in the detection environment according to the environmental noise power and the signal average power; Wherein, when the signal average power is greater than the environmental noise power, it is determined that the target object exists in the detection environment. 4. The method according to claim 1, wherein determining the corresponding range-Doppler information according to the feedback signal comprises: Performing a fast Fourier transform on the fast time signal of the feedback signal after removing DC, to obtain distance dimension information; Performing a fast Fourier transform on the slow-time signal of the feedback signal after removing DC to obtain Doppler information; The distance dimension information is accumulated according to the distance dimension information and the Doppler information to obtain the distance-Doppler information. 5. The method according to claim 4, characterized in that the fast Fourier transform is performed on the fast time signal of the feedback signal after DC removal to obtain the distance dimension information, which comprises: Performing a fast Fourier transform on a fast time signal in multiple slow time dimensions for each frame signal of the feedback signal to obtain a first range image; Performing a fast Fourier transform on the fast time signal of the feedback signal in the frame time dimension to obtain a second range image; The distance dimension information includes the first distance image and the second distance image. 6. The method according to claim 5, characterized in that detecting the height of the target object according to the range-Doppler information and determining the first risk probability of the target object falling comprises: determining, according to the first range image, whether the target object has a movement close to the ground; In the case of the movement close to the ground, a corresponding falling height is obtained according to the human body height in the posture threshold parameter, wherein the falling height corresponds to the human body height; the human body height is determined according to the position of the high power point in the second range image at the range gate and the installation height of the radar; Determining the highest height of the target object according to the first range image; When the maximum height is less than the falling height, it is determined that the target object has the first risk probability of falling. 7. The method according to claim 5, characterized in that detecting the range gate of the ground of the detection environment where the target object is located according to the range-Doppler information to determine whether there is a human breathing frequency feature, and determining the second risk probability of the target object falling comprises: Performing a fast Fourier transform on the range gate within a first preset height range of the floor according to the second range image to obtain first target frequency information; In a case where the first target frequency information has a respiratory frequency feature, it is determined that the target object has the second risk probability of falling. 8. The method according to claim 5, characterized in that determining envelope graphs of different range gates according to the range-Doppler information, and determining the third risk probability of the target object falling comprises: Determine envelope graphs of different range gates according to the first range image; When the envelope graph does not match the envelope graph of the target object when not falling, it is determined that the target object has the third risk probability of falling, wherein the posture threshold parameter includes the envelope graph of the normal posture of the target object when not falling. 9. The method according to claim 5, characterized in that determining the fourth risk probability of the target object falling according to whether the distance-Doppler information has a signal feature of weak breathing comprises: Performing a fast Fourier transform on a range gate within a second preset height range according to the second range image to obtain second target frequency information, wherein the second preset height range is higher than the first preset height range; In a case where the second target frequency information has a respiratory frequency feature, it is determined that the target object has the fourth risk probability of falling. 10. A fall detection device for a target object, comprising: A receiving module, used to receive a feedback signal generated by the target object reflecting the radar's transmission signal; A first determination module, configured to determine corresponding range-Doppler information according to the feedback signal; A second determination module is used to determine the human body velocity information of the target object through the range-Doppler information; A third determination module is used to determine whether the target object falls according to the distance-Doppler information when the human body speed information exceeds a preset speed threshold; When the human body speed information exceeds a preset speed threshold, determining whether the target object falls according to the distance-Doppler information includes: Detecting the height of the target object according to the range-Doppler information, and determining a first risk probability of the target object falling; Detecting the distance gate of the floor of the detection environment where the target object is located according to the distance-Doppler information to determine whether there is a human breathing frequency feature, and determining a second risk probability of the target object falling; Determine envelope graphs of different range gates according to the range-Doppler information, and determine a third risk probability of the target object falling; determining a fourth risk probability of the target object falling according to whether the distance-Doppler information has a signal feature of weak breathing; Determining a comprehensive probability of the target object falling according to the first risk probability, the second risk probability, the third risk probability and the fourth risk probability, and their corresponding weights; When the comprehensive probability reaches a preset probability threshold, it is determined that the target object has fallen. 11. A processor, characterized in that the processor is used to run a program, wherein the program executes the fall detection method for a target object according to any one of claims 1 to 9 when running. 12. A computer storage medium, characterized in that the computer storage medium includes a stored program, wherein when the program is running, the device where the computer storage medium is located is controlled to execute the fall detection method for a target object according to any one of claims 1 to 9.