CN115131582A - Target recognition method, device and medium based on morphological recognition template matching - Google Patents

Abstract

The present application proposes a target recognition method based on morphological recognition template matching, including: S1, obtaining an original image; S2, converting the original image from RGB format to YCBCR format, and converting the original image in the format Perform binarization processing to obtain a binarized image, and according to the first threshold of YCBCR corresponding to each color, screen out the pixels larger than the corresponding first threshold from the binarized image in turn; S3, determine Whether the distance between the two pixels is greater than the second threshold, if so, the two pixels correspond to different targets, if not, the two pixels are the same target, and a new first is generated. Two thresholds; S4 , matching the binarized image with the binarized template images of various targets, so as to determine the type corresponding to each target existing in the binarized image. This application integrates the identification of the color, quantity and type of the target into one, and the resource usage is small, which is suitable for simple usage scenarios.

Description

Target recognition method, device and medium based on morphological recognition template matching

technical field

The present application relates to the technical field of target recognition, in particular to a target recognition method, device and medium based on morphological recognition template matching.

Background technique

With the development of science and technology in today's society, the target recognition algorithm is gradually becoming mature, and the frame signal processing and recognition of the camera video stream has always been a hot issue. Item recognition is one of the basic functions of intelligent machines. This function has a wide range of application scenarios in both military and civilian applications, such as intelligent video surveillance, unmanned driving, various types of identification, and computer forensics.

There are many different technical methods in the development of computer image recognition, which can be roughly divided into traditional image recognition methods and recognition methods based on the fusion of neural network algorithms. Neural network image recognition technology is a relatively new type of image recognition technology. In neural network image recognition technology, the deep learning model based on the combination of convolutional neural network can be described as a new star in the field of artificial intelligence. In particular, impressive progress has been made in the field of image recognition. Although the image recognition algorithm based on neural network is relatively good in accuracy, it consumes too much hardware resources.

Therefore, the present application is to select the traditional identification method, and use less hardware resources to realize target identification under the condition that the usage scenario is relatively simple.

SUMMARY OF THE INVENTION

In order to solve the above-mentioned technical problems in the prior art, the present application proposes a target recognition method, device and medium based on morphological recognition template matching.

According to the first aspect of the present application, a target recognition method based on morphological recognition template matching is proposed, which includes the following steps:

S1. Obtain the original image;

S2. Convert the original image from RGB format to YCBCR format, and perform binarization processing on the converted original image to obtain a binarized image. According to the first threshold of YCBCR corresponding to various colors, Screen out the pixels larger than the corresponding first threshold in sequence from the binarized image;

S3. Determine whether the distance between the two pixel points is greater than the second threshold. If so, the two pixel points correspond to different targets. If not, the two pixel points are the same target, and generate a new second threshold; and

S4. Match the binarized image with the binarized template images of various targets, so as to determine the type corresponding to each target existing in the binarized image.

Preferably, the steps S2-S4 are implemented on an FPGA development board.

Preferably, in the step S2, each time a pixel is screened out, an enable signal of a corresponding color is correspondingly generated; in the step S3, a plurality of candidate frames arranged in sequence are initialized, and in the first When a pixel point appears, assign a value to the first candidate frame according to the corresponding enable signal, so that a detection frame with a corresponding color is framed at the position of the pixel point in the binarized image.

Preferably, in the step S3, when a new pixel appears, it is judged whether the new pixel is in the detection frame corresponding to any old pixel, if so, the new pixel is the same as the old pixel. The points belong to the same target, and the new pixel point is used as the boundary of the detection frame corresponding to the old pixel point. If not, the next candidate frame is assigned according to the enable signal corresponding to the new pixel point. The binarized image corresponds to a new pixel position and selects a detection frame with a corresponding color; wherein, the second threshold is the difference between each pixel with a detection frame and the upper, lower, left, and right boundaries of the detection frame. distance.

Preferably, the step S4 specifically includes: according to each target determined in the step S3, dividing the binarized image into target images corresponding to each target, and the target image and the various targets The size of the binarized template images is the same, and the target image is matched with the binarized template images of various targets, so as to determine the type corresponding to each target.

Preferably, the generation process of the detection frame specifically includes: taking the left border of the row and the left border of the field as the upper left border points, scanning to generate the detection frame with a size of 500*500 pixels, so that the corresponding pixel points fall on the The center of the detection box.

Preferably, the formula for converting the original image from the RGB format to the YCBCR format in the step S2 is specifically:

Y=0.183R+0.614G+0.062B+16

CB=-0.101R-0.338G+0.439B+128

CR=0.439R-0.399G-0.040B+128.

According to a second aspect of the present application, a target recognition device based on morphological recognition template matching is proposed, including:

Image acquisition module, configure the application to acquire the original image;

The color recognition module is configured to convert the original image from RGB format to YCBCR format, and perform binarization processing on the converted original image to obtain a binarized image. According to the YCBCR corresponding to each color the first threshold of , and sequentially filter out the pixels larger than the corresponding first threshold from the binarized image;

A quantity identification module, configured to judge whether the distance between the two pixel points is greater than the second threshold, if so, the two pixel points correspond to different targets, if not, the two pixel points are the same a target and generate a new second threshold;

The template matching module is configured to match the binarized image with the binarized template images of various targets, so as to determine the type corresponding to each target existing in the binarized image.

Preferably, there are multiple color identification modules, and each of the color identification modules corresponds to screening pixels of one color.

According to a third aspect of the present application, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the template matching based on morphology recognition as described in the first aspect of the present application target recognition method.

This application proposes a target recognition method, device and medium based on morphological recognition template matching. First, the image is processed in RGB to YCBCR format and binarized, and then the YCBCR threshold corresponding to each color is segmented, so as to filter The pixels corresponding to various colors in the image are obtained to realize color recognition; after filtering out the pixels, by judging whether the distance between the two pixels is greater than the threshold, it can be judged whether the new pixel is a new target, and the number of achieved Recognition; finally, the target image of each target obtained by judgment is matched with the pre-placed binarized template images of various targets, so as to realize the type recognition of each target in the image. This application not only integrates the color, quantity, and type recognition of the target, but also uses less resources than the traditional neural network target recognition algorithm, can run on the FPGA development board, and is suitable for various simple scenes to be used.

Description of drawings

The accompanying drawings are included to provide a further understanding of the embodiments and are incorporated into and constitute a part of this specification. The drawings illustrate embodiments and together with the description serve to explain the principles of the application. Other embodiments and many of the intended advantages of the embodiments will be readily recognized as they become better understood by reference to the following detailed description. The elements of the drawings are not necessarily to scale to each other. Like reference numerals designate corresponding similar parts.

1 is a flowchart of a target recognition method based on morphological recognition template matching according to an embodiment of the present application;

2 is a schematic diagram of image binarization according to a specific embodiment of the present application;

3 is a schematic diagram of identifying the number of targets through threshold judgment according to a specific embodiment of the present application;

4 is a schematic diagram of a binarized template image stored in a ROM according to a specific embodiment of the present application;

5 is a schematic diagram of template matching according to a specific embodiment of the present application;

6 is a schematic diagram of physical image detection according to a specific embodiment of the present application;

7 is a target recognition result diagram according to a specific embodiment of the present application;

8 is a schematic diagram of physical image detection according to another specific embodiment of the present application;

9 is a target recognition result diagram according to another specific embodiment of the present application;

10 is a three-layer convolutional neural network framework diagram according to a specific embodiment of the present application;

Fig. 11 is a resource usage diagram of a three-layer convolutional neural network according to a specific embodiment of the present application;

FIG. 12 is a schematic structural diagram of a target recognition apparatus based on morphological recognition template matching according to an embodiment of the present application.

Description of reference numerals: 1. Image acquisition module; 2. Color identification module; 3. Quantity identification module; 4. Module matching module.

Detailed ways

The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application more clear, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only configured to explain the present application, and are not configured to limit the present application. It will be apparent to those skilled in the art that the present application may be practiced without some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present application by illustrating examples of the present application.

It should be noted that, in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any relationship between these entities or operations. any such actual relationship or sequence exists. Moreover, the terms "comprising", "comprising" or any other variation thereof are intended to encompass a non-exclusive inclusion such that a process, method, article or device that includes a list of elements includes not only those elements, but also includes not explicitly listed or other elements inherent to such a process, method, article or apparatus. Without further limitation, an element defined by the phrase "comprises..." does not preclude the presence of additional identical elements in the process, method, article, or device that includes the element.

First of all, it should be noted that the target identification method of the present application is implemented based on an FPGA chip as a hardware carrier. The algorithms used by the current FPGA-related target recognition systems need to integrate color, quantity, and type information. Most of the algorithms are only doing object position tracking and type recognition. Therefore, the purpose of this application is to use FPGA as a hardware carrier. The traditional target recognition method integrates the color, quantity, and type recognition of the target under the condition of less resource usage, which is suitable for some simple usage scenarios, thus broadening the application scenarios of FPGA.

According to the first aspect of the present application, a target recognition method based on morphological recognition template matching is proposed. FIG. 1 shows a flowchart of a target recognition method based on morphological recognition template matching according to an embodiment of the present application. As shown in FIG. 1 , the method includes the following steps:

S1. Obtain an original image.

In a specific embodiment, the 5640 camera of ALINX is used to collect the original image, and the image is stored through DDR3.

S2. Convert the original image from RGB format to YCBCR format, and perform binarization processing on the converted original image to obtain a binarized image. According to the first threshold of YCBCR corresponding to each color, from binarization Pixels larger than the corresponding first threshold are screened out in sequence in the image.

In a specific embodiment, after the original image is acquired, the data is imported into the FPGA development board for processing. The formula for converting the original image from RGB format to YCBCR format is as follows:

Y=0.183R+0.614G+0.062B+16

CB=-0.101R-0.338G+0.439B+128

CR=0.439R-0.399G-0.040B+128

Binarizing the image can better distinguish the pixels and background points in the image, making the image simpler and more able to highlight the outline of the target. Taking fruit identification as an example, the main purpose of identifying fruit colors is to find common fruit colors such as red, yellow, green, purple, tan, etc. In this embodiment, MATLAB is used for image processing, and the YCBCR corresponding to each color is determined. In order to prevent interference between the first thresholds of these colors, the first thresholds can be continuously improved so that the first thresholds are within a range. After determining the first threshold of YCBCR corresponding to each color, by comparing the size, from the top to bottom and left to right of the binarized image, filter out the binarized image that is larger than the corresponding first threshold. Threshold pixels to identify the color of these pixels. FIG. 2 shows a schematic diagram of image binarization according to a specific embodiment of the present application.

In a specific embodiment, each time a pixel is compared with the corresponding first threshold, an enable signal will be generated correspondingly. For example, if the YCBCR value of the pixel is greater than or equal to the corresponding first threshold, the enable signal is 1. If the YCBCR value of the pixel is less than the corresponding first threshold, the enable signal is 0.

S3. Determine whether the distance between the two pixels is greater than the second threshold. If so, the two pixels correspond to different targets. If not, the two pixels are the same target, and a new second threshold is generated.

In a specific embodiment, the pixels selected in step S2 may all correspond to one target, or two or more pixels may correspond to the same target, so by judging the new pixel and the old pixel Whether the distance between the points is greater than the second threshold, if it is greater than the second threshold, it is judged that the new pixel corresponds to a new target, if it is less than the second threshold, it is judged that the new pixel corresponds to the same one as the old pixel Target. The following will explain the content of this part with specific embodiments.

First, initialize multiple candidate frames in order, the number of candidate frames can be set by yourself, and the initial horizontal and vertical coordinates of these candidate frames are (0, 0). When the first pixel is screened out in step S2, the first candidate frame is assigned a value according to the corresponding enable signal, so that a detection frame with a corresponding color is framed at the pixel position of the binarized image. The detection frame is generated by line and field positioning scanning. The specific generation process is as follows: take the left border of the line and the left border of the field as the upper left border point, and scan to generate a detection frame with a size of 500*500 pixels, so that the corresponding pixel points fall on the upper left border. The center of the detection box.

When a new pixel is screened out again in step S2, first determine whether the new pixel is in the detection frame where the old pixel is located (that is, determine whether the distance between the new pixel and the old pixel is smaller than the detection frame). The half-side length is 250 pixels), if it is, it can be judged that the new pixel and the old pixel belong to the same target, then this new pixel is used as the boundary of the detection frame corresponding to the old pixel. In this way, The detection frame of the old pixel point is reduced, so that the target in the binarized image can be selected more accurately; if not, it is judged that the new pixel point corresponds to a new target, and then again according to the corresponding enable The signal assigns a value to the next (that is, the second) candidate frame, so as to select a detection frame with a corresponding color at the position of the new pixel point in the binarized image. And so on, until the judgment of all pixels is completed. Therefore, it is not difficult to understand that the second threshold value will be continuously added according to the judgment result, and the second threshold value is equivalent to the distance between each pixel point with a detection frame and the upper, lower, left, and right boundaries of the detection frame.

Fig. 3 shows a schematic diagram of identifying the number of targets through threshold judgment according to a specific embodiment of the present application. As shown in Fig. 3(a), between the first pixel point (left side) and the second pixel point (right side) The distance between the two pixels is greater than the second threshold, so the two pixels correspond to two targets and both display a detection frame; as shown in Figure 3(b), the difference between the first pixel (left) and the second pixel (right) is The distance between the two pixels is greater than the second threshold. Although the detection frames of the two pixels have overlapping parts, they are still judged to correspond to different targets; as shown in Figure 3(c), the first pixel (right) and the second pixel The distance between the pixels (left) is greater than the second threshold, and the second pixel falls within the detection frame of the first pixel, so the two pixels correspond to the same target, and the second pixel is used as the first pixel. The left border of the detection frame of each pixel, and the detection frame of the first pixel is reduced.

S4: Match the binarized image with the binarized template images of various targets, so as to determine the type corresponding to each target existing in the binarized image.

In a specific embodiment, for the type recognition of fruits, in order to save on-chip resources and improve the recognition speed, it is decided to adopt the morphological recognition method after comprehensively considering the characteristics of the target to be recognized. In short, it is to first identify the fruit categories that can be distinguished by color, such as apples, kiwi fruit, dragon fruit and other fruits with unique colors, which can be judged by color. For the same color, such as bananas and mangoes, take the It is the method of template matching to judge. The so-called template matching is to store the binarized template images of various targets such as bananas and mangoes in the FPGA respectively. After the camera captures the bananas or mangoes, the binarized images are segmented according to each target determined in step S3. A target image corresponding to each target, and the target image is the same size as the binarized template image of various targets, the target image is matched with the binary template image of various targets, and the final matching result is counted. The highest score is the For the corresponding species, this method can be applied to the recognition of other fruits with the same color but different species.

FIG. 4 shows a schematic diagram of a binarized template image stored in a ROM according to a specific embodiment of the present application. As shown in FIG. 4 , the binarized template images of mango and banana are stored. FIG. 5 shows a schematic diagram of template matching according to a specific embodiment of the present application. As shown in FIG. 5 , the one on the left is template information, and the two on the right are collected image information. The collected image information and template information are combined with each other. Operation, count the number of 1, obviously the middle picture has the highest matching degree.

In order to verify the reliability of the target recognition algorithm of the present application, the present application also conducts the following target recognition detection experiments.

Fig. 6 shows a schematic diagram of physical image detection according to a specific embodiment of the present application, and Fig. 7 shows a target recognition result graph according to a specific embodiment of the present application. As shown in Fig. 6 and Fig. 7, there are 3 The first column of data in Figure 7 is the color recognition result, the second column of data is the type recognition result, and the third column of data is the quantity recognition result. According to the detection result data of the latest frame image, the detected color It is GREEN (green), the type is grape (grape), and the number is 3.

FIG. 8 shows a schematic diagram of physical image detection according to another specific embodiment of the present application, and FIG. 9 shows a target recognition result graph according to another specific embodiment of the present application. As shown in FIGS. 8 and 9 , in FIG. 8 There is 1 yellow banana. According to the detection result data of the latest frame image, the detected color is YELLOW (yellow), the type is banana (grape), and the quantity is 1.

It can be seen from the above detection results that the target recognition method of the present application is reliable. And it should be noted that in the third column of data in Figure 7 and Figure 9, the number recognition results of some frame images display ">", "<" or "=", this is because the camera is moving when the image is captured. , some frames cannot capture the target well, so the specific quantity cannot be identified.

In this embodiment, a comparison of resource usage between the target recognition method of the present application and the target recognition algorithm based on neural network in the prior art is also given. In this embodiment, the on-chip resources based on the Tsinghua Unigroup PGL22G development board are only 22K. Fig. 10 shows a framework diagram of a three-layer convolutional neural network according to a specific embodiment of the present application, and Fig. 11 shows a resource usage diagram of a three-layer convolutional neural network according to a specific embodiment of the present application, as shown in Fig. 10. As shown in Figure 11, for a three-layer convolutional neural network, the proportion of FPGA resources consumed when implementing these three-layer neural networks is huge, and this is a ZYNQ development board with large computing resources. It is not only implemented with FPGA, and the proportion of resources used is also high.

To sum up, this application proposes a target recognition method based on morphological recognition template matching. First, the image is processed in RGB to YCBCR format and binarized, and then the YCBCR threshold corresponding to each color is segmented, thereby Screen out the pixels corresponding to various colors in the image to realize color recognition; after screening out the pixels, by judging whether the distance between the two pixels is greater than the threshold, it can be judged whether the new pixel is a new target. Quantity identification; finally, the target image of each target obtained by judgment is matched with the binarized template images of various targets put in in advance, so as to realize the type recognition of each target in the image. This application not only integrates the color, quantity, and type recognition of the target, but also uses less resources than the traditional neural network target recognition algorithm, can run on the FPGA development board, and is suitable for various simple scenes to be used.

According to the second aspect of the present application, a target recognition device based on morphological recognition template matching is proposed, and the target recognition device is constructed based on the above target recognition method. FIG. 12 shows a schematic structural diagram of a target recognition device based on morphological recognition template matching according to an embodiment of the present application. As shown in FIG. 12 , the device includes:

The image acquisition module 1 is configured to acquire the original image.

The color recognition module 2 is configured to convert the original image from RGB format to YCBCR format, and perform binarization processing on the converted original image to obtain a binarized image. Threshold, screen out the pixels larger than the corresponding first threshold in turn from the binarized image.

Quantity identification module 3, configured to judge whether the distance between two pixels is greater than the second threshold, if so, the two pixels correspond to different targets, if not, the two pixels are the same target, and generate New second threshold.

The template matching module 4 is configured to match the binarized image with the binarized template images of various targets, so as to determine the type corresponding to each target existing in the binarized image.

In a specific embodiment, a plurality of color identification modules 2 are provided, and each color identification module 2 stores the first threshold data of YCBCR corresponding to one color, so that each color identification module 2 can filter one color correspondingly of pixels.

According to a third aspect of the present application, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the target recognition based on morphological recognition template matching according to the first aspect of the present application method.

In the embodiments of the present application, it should be understood that the disclosed technical content may be implemented in other manners. The apparatus/system/method embodiments described above are only illustrative, for example, the division of the units may be a logical function division, and in actual implementation, there may be other divisions, such as multiple units or components May be combined or may be integrated into another system, or some features may be omitted, or not implemented. On the other hand, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of units or modules, and may be in electrical or other forms.

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

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

The integrated unit, if implemented in the form of a software functional unit and sold or used as an independent product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of software products in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, and the computer software products are stored in a storage medium , including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. 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 .

It will be apparent to those skilled in the art that various modifications and changes to the embodiments of the present application can be made without departing from the spirit and scope of the present application. In this manner, this application is also intended to cover such modifications and changes if they come within the scope of the claims of this application and their equivalents. The word "comprising" does not exclude the presence of other elements or steps not listed in a claim. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage. Any reference signs in the claims should not be construed as limiting the scope.

Claims (10)

1. a target recognition method based on morphological recognition template matching, is characterized in that, comprises the following steps: S1. Obtain the original image; S2. Convert the original image from RGB format to YCBCR format, and perform binarization processing on the converted original image to obtain a binarized image. According to the first threshold of YCBCR corresponding to various colors, Screen out the pixels larger than the corresponding first threshold in sequence from the binarized image; S3. Determine whether the distance between the two pixel points is greater than the second threshold. If so, the two pixel points correspond to different targets. If not, the two pixel points are the same target, and generate a new second threshold; and S4. Match the binarized image with the binarized template images of various targets, so as to determine the type corresponding to each target existing in the binarized image. 2. The method according to claim 1, wherein the steps S2-S4 are implemented on an FPGA development board. 3. The method according to claim 2, characterized in that, in the step S2, each time a pixel is screened out, an enable signal of a corresponding color is generated correspondingly; in the step S3, initializing multiple There are candidate boxes arranged in sequence, when the first pixel point appears, the first candidate box is assigned according to the corresponding enable signal, so as to select the box at the pixel point of the binarized image. A detection frame with the corresponding color is displayed. 4. method according to claim 3, is characterized in that, in described step S3, when appearing a new pixel point, judge whether new pixel point is in the detection frame corresponding to any old pixel point, If so, the new pixel and the old pixel belong to the same target, and the new pixel is used as the boundary of the detection frame corresponding to the old pixel. The energy signal assigns a value to the next candidate frame, so as to select a detection frame with a corresponding color at the position corresponding to the new pixel point of the binarized image; wherein, the second threshold is for each detection frame with a detection frame. The distance between the pixel point and the upper, lower, left, and right boundaries of the detection frame. 5. The method according to claim 2, wherein the step S4 specifically comprises: dividing the binarized image into target images corresponding to each target according to each target determined in the step S3, And the size of the target image and the binarized template images of various targets is the same, and the target image is matched with the binarized template images of various targets, so as to determine the type corresponding to each target. 6. The method according to claim 4, wherein the generation process of the detection frame specifically comprises: taking the row left border and the field left border as the upper left border point, scanning to generate the detection of the size of 500*500 pixel value frame, so that the corresponding pixel point falls in the center of the detection frame. 7. The method according to claim 2, wherein the formula that the original image is converted from RGB format to YCBCR format in the step S2 is specifically: Y=0.183R+0.614G+0.062B+16 CB=-0.101R-0.338G+0.439B+128 CR=0.439R-0.399G-0.040B+128. 8. A target recognition device based on morphological recognition template matching, characterized in that, comprising: Image acquisition module, configure the application to acquire the original image; The color recognition module is configured to convert the original image from RGB format to YCBCR format, and perform binarization processing on the converted original image to obtain a binarized image. According to the YCBCR corresponding to each color the first threshold of , and sequentially filter out the pixels larger than the corresponding first threshold from the binarized image; A quantity identification module, configured to judge whether the distance between the two pixel points is greater than the second threshold, if so, the two pixel points correspond to different targets, if not, the two pixel points are the same a target and generate a new second threshold; The template matching module is configured to match the binarized image with the binarized template images of various targets, so as to determine the type corresponding to each target existing in the binarized image. 9 . The device according to claim 8 , wherein a plurality of the color identification modules are provided, and each of the color identification modules corresponds to screening pixels of one color. 10 . 10. A computer-readable storage medium storing a computer program which, when executed by a processor, implements the method of any of claims 1-7.

Images