Monitoring picture real-time people flow statistical system based on YOLO-V5s model

Abstract

The invention provides a monitoring picture real-time people flow statistical system based on a YOLO-V5s model, and relates to the technical field of picture monitoring. The system comprises a video reading module, an image preprocessing module, a human body recognition module based on a YOLO-V5s model and a people flow statistics module, wherein the video reading module is used for performing real-time video access and instant frame extraction; the image preprocessing module is used for carrying out pixel size standardization processing on the picture frames acquired by the video reading module; the human body identification module completes all human body calibration of a real-time picture of the monitoring instrument; the people flow statistics module is used for carrying out statistics on the real-time people flow of the monitoring instrument picture. The system provided by the invention is excellent in precision, can accurately recognize all human bodies in the picture of the monitoring instrument, is excellent in real-time performance, can display the result of people flow statistics on a large screen in the first time, is high in automation degree, and can effectively reduce the labor cost, and the human body recognition and quantity statistics processes are fully automatically completed by the system.

Description

Technical Field

[0001] The present invention relates to the technical field of picture monitoring, and in particular to a real-time crowd counting system for monitoring pictures based on a YOLO-V5s model. Background Art

[0002] At present, the real-time crowd counting on the monitoring instrument screen is mainly carried out manually, which will have the following problems: low accuracy: the real-time monitoring video screen has a large amount of information, and the human brain cannot fully capture all the information, so it is easy to lose sight of the big picture, and statistical errors are normal; poor real-time performance: the number of people must be counted before the value can be entered, and this process will waste time; waste of resources: at least one or more staff members must be maintained in front of the monitoring instrument screen, which will cause unnecessary loss of manpower costs. Summary of the invention

[0003] The purpose of the present invention is to provide a real-time crowd counting system for monitoring screens based on the YOLO-V5s model to solve the problem of real-time crowd counting in monitoring screens, in view of the above-mentioned deficiencies in the prior art.

[0004] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0005] The present invention provides a real-time crowd counting system based on a monitoring screen of a YOLO-V5s model, the system comprising: a video reading module, an image preprocessing module, a human body recognition module based on a YOLO-V5s model, and a crowd counting module.

[0006] The video reading module is used for real-time video access and instant frame extraction to convert the original video format output by the monitoring instrument into picture frames in image format;

[0007] The image preprocessing module is used to perform pixel size normalization processing on the picture frames obtained by the video reading module. The image preprocessing module is also used to perform Gaussian filtering noise reduction and brightness equalization processing on the image data;

[0008] The human body recognition module based on the YOLO-V5s model uses the YOLO-V5s model to perform human body calibration on the image processed by the image preprocessing module by outputting the filtering function, the logic judgment function and refreshing the recognition information, so as to complete the human body calibration of the real-time image of the monitoring instrument;

[0009] The crowd counting module is used to count the real-time crowd flow on the monitoring instrument screen based on the images with all human body calibration completed, and display the statistical results on the front large screen until the monitoring instrument stops operating.

[0010] Optionally, the video reading module is specifically used to: adapt and filter the communication protocol according to the specific conditions of the deployment location of the monitoring instrument, and then transmit the real-time video information to the specified directory of the local server so that it can be directly read and processed by the server hardware; by intercepting the current screen frame of real-time monitoring, it is converted from video format to jpg image format so that it can be processed by a preset computer vision algorithm.

[0011] Optionally, the image preprocessing module is specifically used to: standardize the pixel size of the current picture frame to a 640*640*3 format, and set its color mode to 16 channels; use a Gaussian low-pass filter in the time domain, and set the filter to a preset field size and a preset standard deviation to remove noise interference during real-time monitoring video transmission; convert the image from RGB coordinates to HSV coordinates, and then use a local histogram equalization algorithm on the V coordinate information to eliminate some uneven lighting interference.

[0012] Optionally, the human body recognition module based on the YOLO-V5s model is specifically used to: use the YOLO-V5s model to perform initial calibration on the image processed by the image preprocessing module; then use the output filter function to calculate the probability of each object being judged as a human body, strengthen the model's recognition function for the human body, and abandon the recognition of other objects; then use the logical judgment function to set the judgment threshold according to the actual conditions of the monitoring area so that the human body recognition result conforms to the actual situation; finally, refresh the recognition information, save the calibration results of the current round of monitoring to the back end, and complete the statistics of real-time human flow in the area. After a new round of model recognition process is started, delete the calibration results of the previous round and retain the calibration results of the new round, so that server resources are not consumed extra.

[0013] Optionally, the crowd counting module is specifically used to: based on the results of the human body recognition module, count the number of human body calibration frames to complete the current population count in the monitoring area; first transmit the population information to the command line, and use the logic verification algorithm to complete the information extraction, and finally display the statistical result information on the front-end large screen.

[0014] The beneficial effects of the present invention include:

[0015] The real-time crowd counting system based on the monitoring screen of the YOLO-V5s model provided by the present invention comprises: a video reading module, an image preprocessing module, a human body recognition module based on the YOLO-V5s model and a crowd counting module, wherein the video reading module is used for real-time video access and instant frame extraction, so as to convert the original video format output by the monitoring instrument into a picture frame in an image format; the image preprocessing module is used for performing pixel size normalization processing on the picture frame acquired by the video reading module, and the image preprocessing module is also used for performing Gaussian filtering noise reduction and brightness equalization processing on the image data; the human body recognition module based on the YOLO-V5s model utilizes the YOLO-V5s model, outputs a filter function, a logic judgment function and refreshes the recognition information, and performs human body calibration on the image processed by the image preprocessing module, so as to complete the calibration of all human bodies in the real-time picture of the monitoring instrument; the crowd counting module is used for counting the real-time crowd flow in the picture of the monitoring instrument based on the image after all human body calibration is completed, and displays the statistical results on the front large screen until the monitoring instrument stops operating. The system proposed by the present invention has excellent accuracy. Except for a few special cases such as being greatly blocked, all human bodies in the monitoring instrument screen can be accurately identified and counted. It has excellent real-time performance, can display the results of human flow statistics on the large screen in the first time, and can complete the update of statistical quantity at a high frequency. It has a high degree of automation, and the human body identification and quantity counting processes are fully automatically completed by the system, which can effectively reduce labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 A schematic diagram of the workflow of a video reading module provided by an embodiment of the present invention is shown;

[0018] Figure 2 A schematic diagram of the workflow of the image preprocessing module provided in an embodiment of the present invention is shown;

[0019] Figure 3 A schematic diagram of the workflow of a human body recognition module based on the YOLO-V5s model provided in an embodiment of the present invention is shown;

[0020] Figure 4 A schematic diagram of the workflow of a crowd counting module provided in an embodiment of the present invention is shown;

[0021] Figure 5A schematic diagram of the workflow of a real-time crowd counting system for monitoring images based on a YOLO-V5s model provided in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] At present, the real-time crowd counting is mainly done manually on the monitoring instrument screen, which will have the following problems: Low accuracy: The real-time monitoring video screen contains a large amount of information. The human brain cannot fully capture all the information and is prone to lose sight of the big picture. It is normal for statistical errors to occur. Poor real-time performance: The number of people must be counted before entering the value, which wastes time. Waste of resources: At least one or more staff members must be maintained in front of the monitoring instrument screen, which will cause unnecessary human cost losses.

[0024] The YOLO-V5s model is a high-level convolutional neural network developed by Glenn Jocher and others. Its main purpose is to locate and identify specific objects in the image. The model structure can be divided into the following four layers according to the order of function and processing: 1. Input layer: perform data enhancement, adaptive anchor box calculation and image scaling to complete image standardization. 2. Backbone layer: extract image object information through the convolutional network to form a feature pyramid. 3. Neck layer: multi-scale feature fusion, mixing and combination of feature maps. 4. Prediction layer: complete the final regression prediction based on comprehensive features. Through the four layers, the prediction box of the target recognition type object in the image can be obtained, and the target recognition task is finally completed. However, the original YOLO-V5s model cannot be directly applied to the real-time crowd counting task of the monitoring instrument screen. This is because the original model can only calibrate the prediction box, but cannot complete the statistics of its number; in addition, when the original YOLO-V5s model is directly connected to the monitoring instrument for data, there will be multiple problems such as incomplete compatibility of the original screen data format and too long calibration and statistical recognition time. Therefore, the YOLO-V5s model needs to be improved to complete the target task.

[0025] This paper proposes a real-time crowd counting system based on the YOLO-V5s model in monitoring instrument images to solve the three problems of accuracy, real-time performance, and resource waste in manual methods. By providing software-level support for video monitoring instruments, it can play a corresponding role in multiple real scenarios such as stampede accident prevention and factory area monitoring.

[0026] Figure 1 A schematic diagram of the workflow of a video reading module provided by an embodiment of the present invention is shown; Figure 2 A schematic diagram of the workflow of the image preprocessing module provided in an embodiment of the present invention is shown; Figure 3 A schematic diagram of the workflow of a human body recognition module based on the YOLO-V5s model provided in an embodiment of the present invention is shown; Figure 4 A schematic diagram of the workflow of a crowd counting module provided in an embodiment of the present invention is shown; Figure 5 A schematic diagram of the workflow of a real-time crowd counting system for monitoring images based on a YOLO-V5s model provided in an embodiment of the present invention is shown.

[0027] The following will refer to Figures 1 to 5 The present invention is described in detail.

[0028] The embodiment of the present invention provides a real-time crowd counting system for monitoring screens based on a YOLO-V5s model. The system comprises: a video reading module, an image preprocessing module, a human body recognition module based on a YOLO-V5s model and a crowd counting module. The video reading module is used for real-time video access and instant frame extraction to convert the original video format output by a monitoring instrument into a picture frame in an image format; the image preprocessing module is used for performing pixel size normalization processing on the picture frame acquired by the video reading module, and the image preprocessing module is also used for performing Gaussian filtering noise reduction and brightness equalization processing on the image data; the human body recognition module based on the YOLO-V5s model utilizes the YOLO-V5s model to perform human body calibration on the image processed by the image preprocessing module by outputting a filter function, a logic judgment function and refreshing identification information to complete all human body calibration of the real-time picture of the monitoring instrument; the human body counting module is used for counting the real-time crowd flow of the monitoring instrument picture based on the image after all human body calibration is completed, and displaying the statistical results on a front-end large screen until the monitoring instrument stops operating.

[0029] like Figure 1As shown, the video reading module is specifically used for: 1. Real-time video access: According to the specific conditions of the deployment location of the monitoring instrument, multiple types of communication protocols such as HTTP and ONVIF are adapted and screened, and then the real-time video information is transmitted to the specified directory of the local server so that it can be directly read and processed by the server hardware; 2. Real-time frame extraction: By intercepting the current screen frame of real-time monitoring, it is converted from common video formats such as sdv and mp4 to jpg image format, so that it can be processed by the preset computer vision algorithm.

[0030] like Figure 2 As shown, the image preprocessing module is specifically used for: 1. Pixel size normalization: standardize the pixel size of the current picture frame to a uniform format of 640*640*3, and set its color mode to 16 channels; for the original image with larger pixels, the mean algorithm is used to complete the image compression; and for the original image with smaller pixels, the interpolation algorithm is used to complete the image expansion. 2. Gaussian filtering noise reduction: use a Gaussian low-pass filter in the time domain, and set the filter to a preset domain size and a preset standard deviation to remove noise interference during the real-time monitoring video transmission process, so that the original quality of the image data is improved; 3. Brightness balance: first convert the image from RGB coordinates to HSV coordinates, and then use a local histogram equalization algorithm for the V coordinate information to eliminate some uneven lighting interference, so that the information in the dark areas of the image data is displayed, and the light intensity in the bright areas is suppressed.

[0031] like Figure 3 As shown in the figure, the human recognition module based on the YOLO-V5s model is specifically used for: using the YOLO-V5s model to perform initial calibration on the image processed by the image preprocessing module; then using the output filtering function to calculate the probability of each object being judged as a human body, streamlining the structure of the output layer, strengthening the model's recognition function for the human body, abandoning the recognition of other objects, and calculating the probability of each object being judged as a human body; so that the method reduces the computational consumption required by the model and speeds up the running speed of the model without affecting the accuracy of the human body; then using the logical judgment function, according to the actual conditions of the monitoring area, setting a suitable discrimination threshold so that the human body recognition result is as consistent as possible with the actual situation; wherein, when the probability that the model judges the calibration frame as a person is greater than this threshold, the type of the frame is judged as a person, and the frame is retained; when the probability that the model judges the calibration frame as a person is less than this threshold, the type of the frame is judged as non-human, and the frame is deleted. Finally, the identification information is refreshed and the calibration results of the current round of monitoring are saved to the back end to complete the statistics of real-time pedestrian flow in the area. When a new round of model recognition process starts, the calibration results of the previous round are deleted and the calibration results of the new round are retained, so that server resources are not consumed extra.

[0032] like Figure 4As shown, the crowd counting module is specifically used for: 1. Real-time crowd counting function: based on the results of human body recognition module, count the number of human body calibration frames and complete the current number of people in the monitoring area; 2. Front-end and back-end docking: first transmit the number of people information to the command line, and use the logic verification algorithm to complete the information extraction, and finally display the statistical result information on the front-end large screen.

[0033] like Figure 5 As shown, when the system is working, first, use the video reading module to convert the original video format into a processable image format; then, use the image preprocessing module to complete the image format normalization and improve its quality; then, use the human body recognition module based on the YOLO-V5s model to complete the calibration of all human bodies in the real-time picture of the monitoring instrument; finally, use the crowd counting module to complete the real-time crowd counting of the monitoring instrument picture, and let the numerical result be displayed on the front end until the monitoring instrument stops operating. In practical applications, find a suitable viewing angle position of looking down or looking straight, and deploy monitoring equipment that can adapt to various transmission protocols in the target monitoring area. On a server where the hardware configuration allows, deploy the video reading module, image preprocessing module, human body recognition module based on the YOLO-V5s model, and crowd counting module, and then install the above modules according to Figure 5 Start the monitoring instrument and connect the code module deployed on the server to officially run the real-time crowd counting system based on the YOLO-V5s model; and regularly output the real-time number of people on the monitoring screen on the large screen until the monitoring instrument stops operating.

[0034] In summary, compared with the prior art, the system proposed by the present invention has excellent accuracy: except for a few special cases such as being largely blocked, all human bodies in the monitoring instrument screen can be accurately identified and counted. Excellent real-time performance: the results of the crowd counting can be displayed on the large screen in the first time, and the statistical quantity can be updated at a high frequency. High degree of automation: the human body recognition and quantity counting processes are all completed automatically by the system, which can effectively reduce labor costs.

[0035] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable ordinary technicians in the field to understand the content of the present invention and implement it. They cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A real-time crowd counting system based on the YOLO-V5s model, characterized in that: The system includes: a video reading module, an image preprocessing module, a human recognition module based on the YOLO-V5s model, and a crowd counting module. The video reading module is used for real-time video access and instant frame extraction to convert the original video format output by the monitoring instrument into picture frames in image format; The image preprocessing module is used to perform pixel size normalization processing on the picture frame obtained by the video reading module, and the image preprocessing module is also used to perform Gaussian filtering noise reduction and brightness equalization processing on the image data; The human body recognition module based on the YOLO-V5s model uses the YOLO-V5s model to perform human body calibration on the image processed by the image preprocessing module by outputting a filtering function, a logic decision function, and refreshing the recognition information, so as to complete the calibration of all human bodies in the real-time picture of the monitoring instrument; The crowd counting module is used to count the real-time crowd flow on the monitoring instrument screen based on the images with all human body calibration completed, and display the statistical results on the front large screen until the monitoring instrument stops operating.

2. According to claim 1, the real-time crowd counting system based on the monitoring screen of the YOLO-V5s model is characterized in that: The video reading module is specifically used to: adapt and screen the communication protocol according to the specific conditions of the deployment location of the monitoring instrument, and then transmit the real-time video information to the specified directory of the local server so that it can be directly read and processed by the server hardware; by intercepting the current picture frame of real-time monitoring, it is converted from video format to jpg image format so that it can be processed by a preset computer vision algorithm.

3. The real-time crowd counting system based on the YOLO-V5s model in monitoring images according to claim 1 is characterized in that: The image preprocessing module is specifically used to: standardize the pixel size of the current picture frame to a 640*640*3 format, and set its color mode to 16 channels; use a Gaussian low-pass filter in the time domain, and set the filter to a preset field size and a preset standard deviation to remove noise interference during real-time monitoring video transmission; convert the image from RGB coordinates to HSV coordinates, and then use a local histogram equalization algorithm on the V coordinate information to eliminate some uneven lighting interference.

4. The real-time crowd counting system based on the monitoring screen of the YOLO-V5s model according to claim 1 is characterized in that: The human body recognition module based on the YOLO-V5s model is specifically used to: use the YOLO-V5s model to initially calibrate the image processed by the image preprocessing module; then use the output filter function to respectively calculate the probability of each object being determined as a human body, strengthen the model's recognition function for the human body, and abandon the recognition of other objects; then use the logical judgment function to set the judgment threshold according to the actual conditions of the monitoring area so that the human body recognition result conforms to the actual situation; Finally, the identification information is refreshed and the calibration results of the current round of monitoring are saved to the back end to complete the statistics of real-time pedestrian flow in the area. When a new round of model recognition process starts, the calibration results of the previous round are deleted and the calibration results of the new round are retained, so that server resources are not consumed extra.

5. The real-time crowd counting system based on the monitoring screen of the YOLO-V5s model according to claim 1 is characterized in that: The crowd counting module is specifically used to: count the number of human body calibration frames based on the results of the human body recognition module to complete the current population statistics in the monitoring area; first transmit the population information to the command line, and use the logic verification algorithm to complete the information extraction, and finally display the statistical result information on the front-end large screen.

Images