CN105139425B - A kind of demographic method and device - Google Patents

Abstract

The application provides a kind of demographic method and device, this method comprises: extracting moving foreground object by carrying out Target Segmentation to current frame image detection zone；Detect the head and shoulder feature frame in the moving foreground object；Judge whether the head and shoulder feature frame in the moving foreground object meets demographics trigger condition；When the head and shoulder feature frame meets the demographics trigger condition, demographics are carried out according to the head and shoulder feature frame.Feature detection time can be shortened by the application, reduce feature false detection rate, lifting feature detection effect, to promote demographics efficiency and accuracy rate.

Description

Method and device for counting people

technical field

The present application relates to the technical field of image processing, and in particular to a method and device for counting people.

Background technique

Real-time people counting systems are deployed in many public places (such as shopping malls, supermarkets, parks, etc.), so that managers can grasp the real-time passenger flow situation, take necessary diversion measures, and prevent dangerous events such as stampede due to too many people.

The current people counting method is mainly based on video detection. Cameras are installed at the entrances and exits of public places to collect video, and then the collected images are analyzed to finally count the number of people in the place. For example, the method of background modeling plus feature library matching is used for pedestrian detection, or multiple head classifiers are used for head detection. However, the above-mentioned people counting methods generally have problems such as low statistical accuracy and low processing efficiency.

Contents of the invention

In view of this, the present application provides a method and device for counting people.

Specifically, this application is achieved through the following technical solutions:

This application provides a method for counting people, which method includes:

Extract the moving foreground target by performing target segmentation on the current frame image detection area;

Detecting the head and shoulders feature frame in the moving foreground target;

Judging whether the head and shoulders feature frame in the moving foreground target satisfies the people counting trigger condition;

When the head and shoulders feature frame satisfies the people counting trigger condition, the people counting is performed according to the head and shoulders feature frame.

The present application also provides a people counting device, which includes:

An extraction unit is used to extract a moving foreground target by performing target segmentation on the current frame image detection area;

A detection unit, configured to detect the head and shoulders feature frame in the moving foreground target;

A judging unit, configured to judge whether the head and shoulders feature frame in the moving foreground object satisfies the trigger condition for people counting;

A counting unit, configured to count people according to the head and shoulders feature frame when the head and shoulders feature frame satisfies the people counting trigger condition.

It can be seen from the above description that the present application can shorten the feature detection time, reduce the feature false detection rate, improve the feature detection effect, and thus improve the efficiency and accuracy of people counting.

Description of drawings

Fig. 1 is a schematic diagram of an application scenario shown in an exemplary embodiment of the present application;

Fig. 2 is a flow chart of a people counting method shown in an exemplary embodiment of the present application;

Fig. 3 is a flow chart of extracting moving foreground objects shown in an exemplary embodiment of the present application;

Fig. 4 is a schematic diagram of the basic hardware structure of a device where a people counting device is shown in an exemplary embodiment of the present application;

Fig. 5 is a schematic structural diagram of a people counting device shown in an exemplary embodiment of the present application.

Detailed ways

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numerals in different drawings refer to the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with aspects of the present application as recited in the appended claims.

The terminology used in this application is for the purpose of describing particular embodiments only, and is not intended to limit the application. As used in this application and the appended claims, the singular forms "a", "the", and "the" are intended to include the plural forms as well, unless the context clearly dictates otherwise. It should also be understood that the term "and/or" as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

It should be understood that although the terms first, second, third, etc. may be used in this application to describe various information, the information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of the present application, first information may also be called second information, and similarly, second information may also be called first information. Depending on the context, the word "if" as used herein may be interpreted as "at" or "when" or "in response to a determination."

Real-time people counting systems are deployed in many public places (such as shopping malls, supermarkets, parks, etc.), so that managers can grasp the real-time passenger flow situation, take necessary diversion measures, and prevent dangerous events such as stampede due to too many people.

The current people counting method is mainly based on video detection. Cameras are installed at the entrances and exits of public places to collect video, and then the collected images are analyzed to finally count the number of people in the place.

Solution 1 of the prior art uses a background modeling algorithm to extract the foreground, then uses a trained feature library to detect pedestrians in the extracted foreground area, and finally tracks and counts the detected pedestrians. However, this statistical method has missed detection in the case of umbrellas and people blocking each other; and in the case of no foreground object segmentation, pedestrian detection takes a long time, resulting in low efficiency of people counting.

In the second prior art solution, a plurality of head classifiers are used for detection, and the head detection rate is high. However, this detection method also has the problem of time-consuming. When it is applied to cameras for real-time detection, it cannot guarantee that every frame of image will be processed, and there is a risk of frame loss, resulting in a decrease in the accuracy of people counting.

In view of the above problems, the embodiment of the present application proposes a people counting method, which combines the detected head and shoulders features with the extracted moving foreground objects to perform people counting, so as to reduce the probability of missed detection; In the process, the method of target segmentation is adopted to reduce the calculation amount of subsequent head and shoulder feature detection and improve the efficiency of people counting.

Referring to FIG. 1 , it is a schematic diagram of a preferred application scenario of the present application. In this application scenario, the camera is installed vertically or nearly vertically, and the viewing angle α of the camera ranges from 65 degrees to 90 degrees. In this application scenario, the situation of people occlusion is less, and the people counting method of the embodiment of the present application is adopted, and the counting accuracy is higher and the speed is faster.

Referring to FIG. 2 , it is a flowchart of an embodiment of the applicant's method for counting the number of people, and this embodiment describes the process of counting the number of people.

Step 201, extracting a moving foreground object by performing object segmentation on the detection area of the current frame image.

The monitoring field of view of the camera is large, and there is no need to process the entire frame of images when extracting moving foreground objects. Usually the key monitoring area of the camera is located in the central area of the image. For example, when the camera monitors the subway gate, although the camera has a large field of view, the really effective area during image processing is only the image area where the subway gate is located. When the camera is installed When the angle and angle are fixed, the position of the effective area in the image is also determined. Therefore, the embodiment of the present application only extracts the moving foreground object from the detection area (equivalent to the aforementioned effective area) in the current frame image, so as to narrow the image processing range and improve the extraction efficiency of the moving foreground object.

Referring to Fig. 3, it is a flowchart for extracting moving foreground objects of the present application, and the specific description is as follows:

Step 2011, acquire the foreground image of the detection area of the current frame image. For example, the mixed Gaussian model can be used for foreground image extraction, and the Gaussian model background can be updated in real time by performing multi-Gaussian modeling on the monitoring scene, thereby improving the accuracy of foreground image extraction.

Step 2012, obtain the foreground target frame by post-processing the foreground image. The post-processing may include median filtering, expansion operation, and regional connection processing, etc. After post-processing, the foreground target frame is obtained. The foreground target frame is the approximate range of the moving foreground target in the detection area. Detection range of foreground objects.

Step 2013, calculating the frame difference map between the current frame image detection area and the previous frame image detection area.

Step 2014, acquire the frame difference edge texture map of the detection area frame difference map. For example, the frame difference edge texture map can be obtained through Sobel (Sobel) processing.

Step 2015, on the obtained frame difference edge texture map, perform horizontal projection and vertical projection on the image area corresponding to the foreground target frame.

Step 2016, acquiring the horizontal projection histogram and the vertical projection histogram generated after the projection.

Step 2017, perform target segmentation on the frame difference edge texture image within the range of the foreground target frame according to the horizontal projection histogram and the vertical projection histogram to obtain a moving foreground target.

First, the processing priorities of the horizontally projected histogram and the vertically projected histogram are calculated separately.

The processing priority calculation method of the horizontal projection histogram and the vertical projection histogram is the same, specifically:

Formula 1)

Among them, u _x is the processing priority; x _i is the number of rows or columns whose projection value is i; ω _i is the weighting coefficient; n is the preset projection threshold.

Wherein, the projection threshold n can select a small value (for example, n=5) according to the experimental data, and when the projection value is less than or equal to the projection threshold, it means that the image region corresponding to the projection value is the background. The smaller the projection value, the more likely the corresponding image area is the background. Therefore, when setting the weighting coefficient ω _i , the smaller the projection value is, the larger the weighting coefficient is set. In this embodiment of the present application, the size of the background area in the projection histogram is reflected by the level of the processing priority, wherein the background area corresponding to the projection histogram with a higher processing priority is larger, and vice versa.

After calculating the processing priority of the horizontal projection histogram and the processing priority of the vertical projection histogram, first, select the projection histogram with high processing priority to perform target segmentation on the frame difference edge texture image within the range of the current foreground target frame, Then, the projection histogram with a low processing priority is selected to perform target segmentation on the frame difference edge texture image after being segmented by the projection histogram with a high processing priority. For example, assuming that the processing priority of the horizontal projection histogram is higher than that of the vertical projection histogram, object segmentation is first performed according to the horizontal projection histogram, and then object segmentation is performed according to the vertical projection histogram.

The target segmentation method of the projection histogram in different directions is the same, specifically, calculate the segmentation cumulative value of the currently selected projection histogram according to the target segmentation algorithm, wherein the target segmentation algorithm can be:

Formula (2)

Formula (3)

Among them, ω _j is the weighting coefficient, which is a positive integer; n is the projection threshold; y _j is the projection value of the jth row or column; f(ω _j ) is the weighting coefficient with positive and negative directions; m ₁ and m ₂ are row or column, and m ₂ >m ₁ ; T _y is the accumulated value of division. As mentioned above, the projection threshold n can take a small value, and the image region whose projection value is less than or equal to the projection threshold n is used as the background.

Select different m ₁ and m ₂ to calculate the split accumulation value by traversing the rows (if the currently selected projection histogram is a horizontal projection histogram) or columns (if the currently selected projection histogram is a vertical projection histogram) of the projection histogram _Ty .

When the segmentation accumulation value T _y is greater than or equal to the preset segmentation threshold T, and When it is smaller than the segmentation threshold T, it is confirmed that m ₁ and m ₂ are a group of target segmentation lines, and the image area between the target segmentation lines m ₁ and m ₂ is the background. After the confirmation of all the target segmentation lines of the current projection histogram is completed, the segmented background area can be eliminated to reduce the computation load of subsequent processing. This is also the reason why the embodiment of the present application performs target segmentation according to the processing priority from high to low, that is, first process from the direction containing more background, remove most of the background, and then process from the direction containing relatively less background, remove remaining background, improving the overall efficiency of object segmentation.

After the above object segmentation is completed, a plurality of image blocks containing moving objects can be obtained. Hereinafter, the image blocks containing moving objects are referred to as moving foreground objects for short, and the moving foreground objects are used for subsequent head and shoulder feature detection. It can be seen that the present application can narrow the detection range of subsequent head and shoulders features through target segmentation, thereby reducing the false detection rate of features and improving the detection efficiency of features.

Step 202, detecting the head and shoulders feature frame in the moving foreground object.

Use the existing feature detection algorithm (for example, Adaboost algorithm) to detect the head and shoulders features in the moving foreground target. When training the classifier, the head and shoulders images with the same angle, close size and similar features can be selected as positive samples to improve the speed of feature detection.

By performing head and shoulder detection on moving foreground objects, multiple head and shoulder feature boxes can be detected, and each head and shoulder feature box represents a person. However, due to errors in feature detection, multiple head and shoulder feature boxes may be detected for the same person. If special processing is not performed, the error in subsequent population counts will be large.

The embodiment of the present application utilizes the characteristic that feature frames of the same target overlap each other, and fuses multiple overlapping head and shoulder feature frames into one head and shoulder feature frame for subsequent feature tracking and matching.

The fusion processing process is as follows: select two head and shoulders feature boxes that have not been fused with each other from the currently existing multiple head and shoulders feature boxes, and obtain the area of the two head and shoulders feature boxes and the area of the two head and shoulders feature boxes respectively. intersection area. According to the area and intersection area of the two head and shoulders feature boxes, it is judged whether the two head and shoulders feature boxes meet the fusion conditions. When the two head and shoulders feature boxes meet the fusion conditions, the minimum circumscribed rectangle of the two head and shoulders feature boxes is used as the new head Shoulder feature box; when two head and shoulder feature boxes do not meet the fusion conditions, they exist as independent head and shoulder feature boxes.

Determine whether fusion processing has been performed between any two currently existing head and shoulder feature frames, and if so, stop the fusion processing. The currently existing head and shoulder feature frames are all independent head and shoulder feature frames, that is, each head and shoulder feature frame can be considered The box corresponds to a person; if not, return to continue the fusion process.

Among them, the process of judging whether the two head and shoulder feature boxes meet the fusion conditions is as follows:

Firstly, it is judged whether the intersection area Area_Over of two head and shoulder feature frames is greater than the standard head and shoulder feature frame area Area_S multiplied by a preset area percentage threshold θ (for example, θ=50%). Wherein, the area of the standard head and shoulders feature frame Area_S is the square of the preset head and shoulders standard width Width in the current application scenario.

When the intersection area Area_Over is less than or equal to the area Area_S of the standard head and shoulders feature box multiplied by the preset area percentage threshold θ, it means that the intersection area of the two head and shoulders feature boxes is small or completely independent. Therefore, it is confirmed that the two head and shoulders feature boxes do not satisfy Fusion conditions.

When the intersection area Area_Over is greater than the area Area_S of the standard head and shoulders feature box multiplied by the preset area percentage threshold θ, it means that the intersection area of the two head and shoulders feature boxes initially meets the fusion condition, and further confirmation is needed.

After the intersection area of the two head and shoulders feature boxes initially meets the fusion condition, the percentage of the intersection area of the two head and shoulders feature boxes is calculated:

Formula (4)

ω _a +ω _b = 1, ω _a > 0, ω _b > 0 Formula (5)

Among them, ω _a is the weighting coefficient of the head and shoulders feature box A; ω _b is the weighting coefficient of the head and shoulders feature box B; Area_Over is the intersection area of the head and shoulders feature box A and the head and shoulders feature box B; Area_A is the head and shoulders feature box A area; Area_B is the area of the head and shoulders feature box B; p is the cross area percentage of the head and shoulders feature box A and the head and shoulders feature box B.

When the intersection area percentage p of the two head and shoulder feature boxes is greater than the preset area percentage threshold θ, it is confirmed that the two head and shoulder feature boxes meet the fusion condition. When the intersection area percentage p of the two head and shoulder feature boxes is less than or equal to the preset area percentage threshold θ, it is confirmed that the two head and shoulder feature boxes do not meet the fusion condition.

Step 203, judging whether the head and shoulders feature in the moving foreground object satisfies the trigger condition of people counting.

After the head and shoulders feature frame detection in step 202 is completed, target matching and trajectory tracking are performed on the detected head and shoulders feature frame.

The matching process of the head and shoulders feature box is as follows: obtain the area and position of the head and shoulders feature box in the current frame image and the previous frame image. Suppose, the width of the head and shoulders feature box A in the current frame image is w _a , then the area of the head and shoulders feature box A is The width of the head and shoulders feature box B in the previous frame image is w _b , then the area of the head and shoulders feature box B is The coordinates of the head and shoulders feature box usually adopt the coordinates of the center point of the feature box. Assume that the coordinates of the head and shoulders feature box A are (x _a , y _a ), and the coordinates of the head and shoulders feature box B are (x _b , y _b ).

According to the acquired area and position of the head and shoulders feature box, determine whether the head and shoulders feature box in the current frame image and the previous frame image match, which can be confirmed according to the following formula.

Formula (6)

Formula (7)

Formula (8)

Formula (9)

ω ₁ +ω ₂ =1, ω ₃ ×η+ω ₄ =1, η>1 Formula (10)

Among them, dist(a, b) is the distance between the head and shoulders feature box A and the head and shoulders feature box B; diff_area(a, b) is the area mean square difference between the head and shoulders feature box A and the head and shoulders feature box B; direction means The moving direction of the head and shoulders feature frame A relative to the head and shoulders feature frame B, 0 and 1 represent two opposite moving directions; θ ₁ and θ ₂ are preset distance thresholds in different moving directions; Thr_Direction means selecting ω ₁ and ω ₃ are the weight coefficients of the distance; ω ₂ and ω ₄ are the weight coefficients of the mean square error of the area; η is the importance coefficient, indicating that the distance value is more important; Thr is the matching evaluation value.

In formula (8), two distance thresholds (θ ₁ and θ ₂ ) related to the moving direction are set. This is due to the installation angle of the camera. The moving distances are different, therefore, the embodiments of the present application improve the accuracy of head and shoulders feature matching by setting different distance thresholds in different moving directions.

The matching evaluation value Thr of the head and shoulders feature boxes in two adjacent frames of images is calculated by the above formula, and the matching evaluation value Thr reflects the matching degree of the head and shoulders feature boxes in two adjacent frames of images. A preset matching evaluation threshold δ is acquired, and it is judged whether the matching evaluation value Thr is smaller than the matching evaluation threshold δ. When the matching evaluation value Thr is less than the matching evaluation threshold δ, it is determined that the head and shoulder feature boxes in the current frame image and the previous frame image match; otherwise, it is determined that the head and shoulder feature boxes in the current frame image and the previous frame image do not match, and the current The head and shoulders feature box in the frame image is a new head and shoulders feature box.

From the above head and shoulders feature matching process, it can be seen that the head and shoulders feature matching principle of the embodiment of the present application is: the closer the distance and the closer the area of the head and shoulders feature boxes in adjacent frame images, the higher the matching degree.

After the matching of the head and shoulders features is determined, track the head and shoulders features. Specifically, record the position of the head and shoulders feature box in the current frame image, referred to as the current position; record the position where the head and shoulders feature box first appears in the image detection area, that is, the starting position; accumulate the number of occurrences of the head and shoulders feature box .

After obtaining the trajectory tracking information of the above-mentioned head-and-shoulders feature, it is judged whether the head-shoulders feature satisfies the people counting trigger condition, specifically: judging whether the head-shoulders feature frame is far away from the counting trigger line along the movement direction, wherein the movement direction is head-shoulders The direction of the feature frame from the initial position to the current position, the counting trigger line is a preset line in the detection area; determine whether the inter-frame movement distance of the head and shoulders feature box is greater than or equal to the preset inter-frame movement distance threshold, Wherein, the moving distance between frames is the moving distance of the head and shoulders feature box between the current frame image and the previous frame image.

When the head and shoulders feature box moves away from the counting trigger line along the motion direction, and the moving distance between frames of the head and shoulders feature box is greater than or equal to the preset moving distance threshold between frames, it is confirmed that the head and shoulders feature box meets the people counting trigger condition; otherwise, Confirm that the head and shoulders feature box does not meet the people counting trigger condition.

The following is a further supplementary description of the trigger conditions for the above-mentioned people counting:

Condition 1: The head-and-shoulders feature frame moves away from the counting trigger line along the movement direction.

This trigger condition is applicable to at least the following two scenarios: Scenario 1, assuming that the starting position of the head and shoulders feature frame is above the counting trigger line, and the current position is below the counting trigger line, it means that the movement direction of the head and shoulders feature frame is from Up to down, and has crossed the counting trigger line to move away from the counting trigger line, so people can be counted. Scenario 2, assuming that the starting position of the head and shoulders feature box is below the counting trigger line, the head and shoulders feature box moves in the area below the counting trigger line, and finally leaves from the lower edge of the detection area, then when the head and shoulders feature box moves to the start When the position is below, it can be determined according to the starting position and current position that its movement direction is from top to bottom, and it moves in a direction away from the counting trigger line, and it can also perform people counting.

The second condition is that the inter-frame movement distance of the head and shoulders feature frame is greater than or equal to a preset inter-frame movement distance threshold.

Since there are false detections in the head and shoulders feature frame detection, for example, some static background objects are regarded as human head and shoulders features, so the detected head and shoulders feature frames need to be further screened. In this embodiment of the present application, a threshold of moving distance between frames is preset by taking advantage of the small moving distance of background objects. When the moving distance of the head and shoulders feature frame in two adjacent frames is greater than or equal to the preset moving distance threshold between frames, It is considered that the current head and shoulders feature box is a true and credible head and shoulders feature box.

Step 204, when the head and shoulders feature frame satisfies the people counting trigger condition, perform people counting according to the head and shoulders feature frame.

When step 203 confirms that the head-and-shoulders feature frame satisfies the triggering condition for people counting, count plus one in the moving direction (from the initial position to the current position) of the head-and-shoulders feature frame, and mark the head-and-shoulders feature frame as Head and shoulders feature boxes are counted to avoid double counting.

In addition, in the embodiment of the present application, when the moving foreground object leaves the detection area, it is judged whether the head and shoulders feature boxes in the moving foreground object have not participated in people counting. When none of the head and shoulder feature boxes in the moving foreground object has participated in the people counting, the people counting is performed according to the moving foreground object.

The foregoing description introduces the method of counting people based on head and shoulder features, but there is a certain rate of missed detection in this statistical method. to count.

In this embodiment of the present application, an auxiliary counting method based on moving foreground objects is added on the basis of counting people based on head and shoulder features. The auxiliary counting method judges the counting statistics of the head and shoulders feature frames in the moving foreground object before the moving foreground object is about to leave the current detection area. The moving foreground target performs people counting to reduce the missed detection rate.

Specifically, after the moving foreground object is extracted through step 201, object matching and trajectory tracking are performed on the moving foreground object.

The matching process of the moving foreground object is as follows: the area of the moving foreground object in the current frame image and the previous frame image is obtained, and when the overlapping area of the moving foreground object in the two frame images is greater than the preset overlapping area threshold, the moving foreground is confirmed target match.

After the moving foreground target is successfully matched, track the moving foreground target. Specifically, record the position of the moving foreground object in the current frame image, that is, the current position; record the position where the moving foreground object first appears in the image detection area, that is, the starting position; and accumulate the number of occurrences of the moving foreground object.

When it is confirmed that none of the head and shoulder feature boxes in the moving foreground object has participated in the people counting, the matching tracking results of the moving foreground object are used for the people counting.

First, set the occurrence threshold of moving foreground objects. The threshold of occurrence times is set according to the number of occurrences of the head and shoulders feature box that has currently participated in the counting of people. Specifically, calculate the average number of occurrences of the head and shoulders feature boxes of the first N number of participants, and multiply the average number of occurrences by a preset adjustment coefficient as the threshold for the number of occurrences of the moving foreground target.

Among them, the average number of occurrences of the head and shoulders feature box is always calculated according to the head and shoulders feature box of the latest participant statistics, so that the occurrence threshold calculated according to the average number of occurrences is not a fixed value, but a value that can be determined according to the application The data of real-time changes in the environment can improve the accuracy of people counting. In addition, because when a person passes through the detection area, the number of occurrences of the head and shoulders feature frame will be less than the number of occurrences of the relatively stable moving foreground target. Therefore, this application adds an average number of occurrences of the head and shoulders feature frame An adjustment coefficient, the adjustment coefficient is greater than 1, so that the setting of the threshold of the number of occurrences of the moving foreground object is more reasonable.

After completing the setting of the threshold of the number of appearances of the moving foreground object, it is judged whether the number of appearances of the current moving foreground object is greater than the set threshold of appearances. When the number of occurrences of the moving foreground object is greater than the set occurrence threshold, add one to the number of people in the moving direction of the current moving foreground object (the direction of the moving foreground object from the initial position to the current position).

It can be seen from the above description that this application combines the detected head and shoulder features with the extracted moving foreground target for people counting, which reduces the probability of missed detection of people counting and enhances the scene adaptability of people counting; at the same time, through the target Methods such as segmentation reduce the amount of computation in the process of people counting, so that the people counting method of the present application can be applied to video surveillance equipment with relatively weak processing capabilities such as cameras, and improve the real-time performance and efficiency of people counting.

Corresponding to the embodiments of the aforementioned method for counting people, the present application also provides embodiments of a device for counting people.

The embodiments of the number counting device of the applicant can be applied to image processing equipment. The device embodiments can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by running the corresponding computer program instructions in the memory on the processor of the device where it is located. From the perspective of hardware, as shown in Figure 4, it is a hardware structure diagram of the device where the applicant's number counting device is located. In addition to the processor, other interfaces, and memory shown in Figure 4, the device in the embodiment is located Usually, the device may also include other hardware according to the actual function of the device, which will not be repeated here.

Please refer to FIG. 5 , which is a schematic structural diagram of a people counting device in an embodiment of the present application. The people counting device includes an extraction unit 501, a detection unit 502, a judgment unit 503 and a statistics unit 504, wherein:

An extraction unit 501, configured to extract a moving foreground target by performing target segmentation on the current frame image detection area;

A detection unit 502, configured to detect the head and shoulders feature frame in the moving foreground object;

A judging unit 503, configured to judge whether the head and shoulders feature frame in the moving foreground object satisfies the trigger condition of people counting;

The counting unit 504 is configured to perform counting of people according to the head and shoulders feature frame when the head and shoulders feature frame satisfies the people counting trigger condition.

Further, the extracting unit 501 includes:

The foreground image acquisition module is used to acquire the foreground image of the current frame image detection area;

The foreground target frame acquisition module is used to obtain the foreground target frame by post-processing the foreground image;

A frame difference map calculation module, used to calculate the frame difference map of the current frame image detection area and the previous frame image detection area;

The frame difference texture map acquisition module is used to obtain the frame difference edge texture map of the frame difference map of the detection area;

An image projection module, configured to perform horizontal projection and vertical projection on the image area corresponding to the foreground target frame on the frame difference edge texture map;

A histogram acquisition module, configured to acquire a horizontal projection histogram and a vertical projection histogram generated after projection;

An object segmentation module, configured to perform object segmentation on the frame difference edge texture image within the range of the foreground object frame according to the horizontal projection histogram and the vertical projection histogram to obtain the moving foreground object.

Further, the target segmentation module includes:

A priority calculation submodule, configured to calculate the processing priority of the horizontal projection histogram and the processing priority of the vertical projection histogram respectively;

The target segmentation sub-module is used to select the projection histogram with high processing priority to perform target segmentation on the frame difference edge texture image within the range of the current foreground target frame; select the projection histogram with low processing priority to process the projection with high priority The frame difference edge texture image after histogram segmentation is subjected to object segmentation to obtain several moving foreground objects.

Further, the priority calculation submodule is specifically used for:

The processing priority calculation method of the horizontal projection histogram and the vertical projection histogram is the same, specifically:

in,

u _x is the processing priority;

x _i is the number of rows or columns whose projection value is i;

ω _i is the weighting coefficient;

n is a preset projection threshold.

Further, the target segmentation submodule is specifically used for:

The method of selecting a projection histogram with a high processing priority for target segmentation is the same as the method for selecting a projection histogram with a low processing priority for target segmentation, specifically:

Calculate the segmentation cumulative value of the selected projection histogram according to the target segmentation algorithm, wherein the target segmentation algorithm is:

in,

ω _j is the weighting coefficient and is a positive integer;

n is the projection threshold;

y _j is the projection value of the jth row or column;

f(ω _j ) is the weighting coefficient with positive and negative directions;

m ₁ and m ₂ are rows or columns, and m ₂ >m ₁ ;

T _y is the division accumulation value;

When the segmentation accumulation value T _y is greater than or equal to the preset segmentation threshold T, and When it is smaller than the segmentation threshold T, it is confirmed that m ₁ and m ₂ are target segmentation lines.

Further, the device also includes:

A tracking unit, configured to perform target matching and trajectory tracking on the head and shoulders feature frame after the detection unit 502 detects the head and shoulders feature frame in the moving foreground object;

The recording unit is used to record the current position of the head and shoulders feature frame in the current frame image according to the target matching and trajectory tracking results; record the initial position where the head and shoulders feature frame appears in the image detection area for the first time;

The judging unit 503 is specifically used to judge whether the head and shoulders feature frame is away from the counting trigger line in a moving direction, the moving direction being the direction of the head and shoulders feature frame from the initial position to the current position; Judging whether the inter-frame movement distance of the head and shoulders feature frame is greater than or equal to a preset inter-frame movement distance threshold, the inter-frame movement distance is the distance between the current frame image and the previous frame image of the head and shoulders feature box Moving distance; when the head and shoulders feature frame moves away from the counting trigger line along the movement direction, and the moving distance between frames of the head and shoulders feature frame is greater than or equal to the preset moving distance threshold between frames, determine the head and shoulders feature frame The people counting trigger condition is met; otherwise, it is determined that the head and shoulders feature box does not meet the people counting trigger condition.

For the implementation process of the functions and effects of each unit in the above device, please refer to the implementation process of the corresponding steps in the above method for details, and will not be repeated here.

As for the device embodiment, since it basically corresponds to the method embodiment, for related parts, please refer to the part description of the method embodiment. The device embodiments described above are only illustrative, and 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 it can be distributed to multiple network elements. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this application. It can be understood and implemented by those skilled in the art without creative effort.

The above is only a preferred embodiment of the application, and is not intended to limit the application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the application should be included in the application. within the scope of protection.

Claims (11)

1. A method for people demographics, the method comprising:

acquiring a foreground image of an effective area of a current frame image;

obtaining a foreground target frame by carrying out post-processing on the foreground image;

calculating a frame difference image of the effective area of the current frame image and the effective area of the previous frame image;

acquiring a frame difference edge texture map of the effective area frame difference map;

performing horizontal projection and vertical projection on an image area corresponding to the foreground target frame on the frame difference edge texture map;

acquiring a horizontal projection histogram and a vertical projection histogram generated after projection;

respectively calculating the processing priority of the horizontal projection histogram and the processing priority of the vertical projection histogram;

selecting a projection histogram with high processing priority to perform target segmentation on the frame difference edge texture image within the range of the current foreground target frame; selecting a projection histogram with low processing priority to perform target segmentation on the frame difference edge texture image segmented by the projection histogram with high processing priority so as to obtain a plurality of moving foreground targets;

detecting the motion foreground target, preliminarily obtaining a head-shoulder feature frame, and performing fusion processing on the preliminarily obtained head-shoulder feature frame based on a preset fusion strategy to obtain a target head-shoulder feature frame uniquely corresponding to the motion foreground target;

judging whether the target head and shoulder feature frame meets a people counting triggering condition;

and when the target head and shoulder feature frame meets the people counting triggering condition, counting the people according to the target head and shoulder feature frame.

2. The method of claim 1, wherein said separately calculating processing priorities for said horizontal projection histogram and said vertical projection histogram comprises:

the processing priority calculation method of the horizontal projection histogram and the vertical projection histogram is the same, and specifically comprises the following steps:

wherein,

u_xis a processing priority;

x_ithe number of rows or columns with a projection value of i;

ω_iis a weighting coefficient;

n is a preset projection threshold.

3. The method of claim 1, wherein the projection histogram with high priority for selection processing performs target segmentation on the frame difference edge texture image within the current foreground target frame; selecting a projection histogram with low processing priority to perform target segmentation on the frame difference edge texture image segmented by the projection histogram with high processing priority, wherein the method comprises the following steps:

the method for selecting the projection histogram with the high processing priority for target segmentation is the same as the method for selecting the projection histogram with the low processing priority for target segmentation, and specifically comprises the following steps:

calculating a segmentation accumulation value of the selected projection histogram according to a target segmentation algorithm, wherein the target segmentation algorithm is as follows:

wherein,

ω_jis a weighting coefficient and is a positive integer;

n is a projection threshold;

y_jis the projected value of the jth row or column;

f(ω_j) Weighting coefficients with positive and negative directions;

m₁、m₂is a row or column, and m₂＞m₁；

T_yIs a segmentation accumulated value;

when the segmentation accumulated value T_yIs greater than or equal to a preset segmentation threshold T, andwhen the value is less than the division threshold value T, m is confirmed₁And m₂Is the target parting line.

4. The method according to claim 1, wherein after the detecting the moving foreground object, preliminarily obtaining a head-shoulder feature frame, and performing fusion processing on the preliminarily obtained head-shoulder feature frame based on a preset fusion strategy to obtain a target head-shoulder feature frame uniquely corresponding to the moving foreground object, the method further comprises:

carrying out target matching and track tracking on the target head and shoulder feature frame;

recording the current position of the target head and shoulder feature frame in the current frame image according to the target matching and track tracking results; recording the initial position of the target head-shoulder characteristic frame appearing in the effective region of the image for the first time;

the judging whether the target head and shoulder feature frame meets the triggering condition of people counting comprises the following steps:

judging whether the target head-shoulder characteristic frame is far away from a counting trigger line along a movement direction, wherein the movement direction is the direction from the starting position to the current position of the target head-shoulder characteristic frame;

judging whether the inter-frame moving distance of the target head-shoulder feature frame is greater than or equal to a preset inter-frame moving distance threshold, wherein the inter-frame moving distance is the moving distance of the target head-shoulder feature frame between the current frame image and the previous frame image;

when the target head-shoulder feature frame is far away from a counting trigger line along the movement direction and the inter-frame movement distance of the target head-shoulder feature frame is greater than or equal to a preset inter-frame movement distance threshold, determining that the target head-shoulder feature frame meets the people counting trigger condition; otherwise, determining that the target head and shoulder feature box does not meet the people counting triggering condition.

5. The method of claim 1, wherein the method further comprises:

when the moving foreground target leaves the effective area, judging whether the head and shoulder feature frames in the moving foreground target do not participate in the counting of the number of people;

and when none of the head and shoulder feature frames in the motion foreground target participates in people counting, carrying out people counting according to the motion foreground target.

6. The method as claimed in claim 5, wherein after extracting the moving foreground object by performing object segmentation on the effective region of the current frame image, the method further comprises:

carrying out target matching and track tracking on the moving foreground target;

recording the current position of the moving foreground target in the current frame image according to the target matching and track tracking results; recording the initial position of the moving foreground target appearing in the effective area of the image for the first time; recording the occurrence times of the moving foreground target;

the counting of the number of people according to the moving foreground target comprises the following steps:

setting a threshold value of the occurrence times of the moving foreground target;

and when the occurrence frequency of the moving foreground target is greater than the occurrence frequency threshold, adding one according to the moving direction count of the moving foreground target, wherein the moving direction is the direction from the starting position to the current position of the moving foreground target.

7. The method of claim 6, wherein setting a threshold number of occurrences of moving foreground objects comprises:

calculating the average value of the occurrence times of the head and shoulder characteristic frames of the first N participated people;

and multiplying the average occurrence number by a preset adjusting coefficient to serve as the threshold of the occurrence number of the moving foreground target.

8. A people counting device, the device comprising:

an extraction unit comprising:

the foreground image acquisition module is used for acquiring a foreground image of the effective area of the current frame image;

the foreground target frame acquisition module is used for obtaining a foreground target frame by carrying out post-processing on the foreground image;

the frame difference image calculation module is used for calculating the frame difference images of the effective area of the current frame image and the effective area of the previous frame image;

the frame difference texture map acquisition module is used for acquiring a frame difference edge texture map of the effective area frame difference map;

the image projection module is used for performing horizontal projection and vertical projection on an image area corresponding to the foreground target frame on the frame difference edge texture map;

the histogram acquisition module is used for acquiring a horizontal projection histogram and a vertical projection histogram which are generated after projection;

an object segmentation module comprising:

the priority calculation submodule is used for respectively calculating the processing priority of the horizontal projection histogram and the processing priority of the vertical projection histogram;

the target segmentation submodule is used for selecting a projection histogram with high processing priority to perform target segmentation on the frame difference edge texture image within the range of the current foreground target frame; selecting a projection histogram with low processing priority to perform target segmentation on the frame difference edge texture image segmented by the projection histogram with high processing priority so as to obtain a plurality of moving foreground targets;

the detection unit is used for detecting the motion foreground target, preliminarily obtaining a head-shoulder characteristic frame, and fusing the preliminarily obtained head-shoulder characteristic frame based on a preset fusion strategy to obtain a target head-shoulder characteristic frame uniquely corresponding to the motion foreground target;

the judging unit is used for judging whether the target head and shoulder characteristic frame meets the people counting triggering condition;

and the counting unit is used for counting the number of people according to the target head and shoulder characteristic frame when the target head and shoulder characteristic frame meets the number of people counting triggering condition.

9. The apparatus of claim 8, wherein the priority computation submodule is specifically configured to:

the processing priority calculation method of the horizontal projection histogram and the vertical projection histogram is the same, and specifically comprises the following steps:

wherein,

u_xis a processing priority;

x_ithe number of rows or columns with a projection value of i;

ω_iis a weighting coefficient;

n is a preset projection threshold.

10. The apparatus of claim 8, wherein the target segmentation submodule is specifically configured to:

the method for selecting the projection histogram with the high processing priority for target segmentation is the same as the method for selecting the projection histogram with the low processing priority for target segmentation, and specifically comprises the following steps:

calculating a segmentation accumulation value of the selected projection histogram according to a target segmentation algorithm, wherein the target segmentation algorithm is as follows:

wherein,

ω_jis a weighting coefficient and is a positive integer;

n is a projection threshold;

y_iis the projected value of the jth row or column;

f(ω_j) Weighting coefficients with positive and negative directions;

m₁、m₂is a row or column, and m₂＞m₁；

T_yIs a segmentation accumulated value;

when the segmentation accumulated value T_yIs greater than or equal to a preset segmentation threshold T, andwhen the value is less than the division threshold value T, m is confirmed₁And m₂Is the target parting line.

11. The apparatus of claim 8, wherein the apparatus further comprises:

the tracking unit is used for detecting the movement foreground target by the detection unit, preliminarily obtaining a head-shoulder characteristic frame, carrying out fusion processing on the preliminarily obtained head-shoulder characteristic frame based on a preset fusion strategy, and carrying out target matching and track tracking on the target head-shoulder characteristic frame after obtaining the target head-shoulder characteristic frame uniquely corresponding to the movement foreground target;

the recording unit is used for recording the current position of the target head and shoulder feature frame in the current frame image according to the target matching and track tracking results; recording the initial position of the target head and shoulder feature frame appearing in the image detection area for the first time;

the judging unit is specifically configured to judge whether the target head-shoulder feature frame is far away from a counting trigger line along a moving direction, where the moving direction is a direction from the starting position to the current position of the head-shoulder feature frame; judging whether the inter-frame moving distance of the target head-shoulder feature frame is greater than or equal to a preset inter-frame moving distance threshold, wherein the inter-frame moving distance is the moving distance of the target head-shoulder feature frame between the current frame image and the previous frame image; when the target head-shoulder feature frame is far away from a counting trigger line along the movement direction and the inter-frame movement distance of the target head-shoulder feature frame is greater than or equal to a preset inter-frame movement distance threshold, determining that the target head-shoulder feature frame meets the people counting trigger condition; otherwise, determining that the target head and shoulder feature box does not meet the people counting triggering condition.