Intelligent crowd monitoring system based on deep learning

Abstract

The invention discloses an intelligent crowd monitoring system based on deep learning, and relates to the technical field of intelligent monitoring, and the system comprises a data collection module which collects monitoring video information in real time and generates a monitoring video stream; the video processing module is used for performing target detection on the acquired monitoring video stream and extracting crowd position and quantity information according to a target detection output result; the image recognition module is used for calculating the crowd density, tracking the movement track of the crowd to obtain movement track data, and calculating crowd distribution change data according to the crowd density; the crowd analysis module is used for predicting and analyzing crowd behaviors by using a deep learning model to obtain a predicted congestion path; and the decision adjustment module is used for judging whether the crowd advancing route needs to be dynamically adjusted based on fuzzy reasoning and warning the crowd to avoid the congestion area. The problem that multi-dimensional comprehensive analysis of crowd behaviors, flowing directions and the like is lacked in the prior art is solved, and the crowd monitoring accuracy and the congestion avoiding efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent monitoring technology, and more specifically, to an intelligent crowd monitoring system based on deep learning. Background Art

[0002] With the acceleration of urbanization and the increase in large-scale gatherings and activities, crowd monitoring and management has become an important need in modern society. Traditional monitoring systems, through manual inspections or simple statistical analysis methods, are unable to cope with the complexity and real-time nature of large-scale crowd behavior analysis. The rapid development of deep learning technology, especially convolutional neural networks (CNN) and recurrent neural networks (RNN) models, has provided strong technical support, making intelligent crowd monitoring based on images or video streams possible.

[0003] An intelligent crowd monitoring system based on deep learning is one that uses deep learning technology, especially computer vision and pattern recognition algorithms, to monitor, analyze and manage crowd dynamics in a specific area in real time.

[0004] Deficiencies of existing technologies:

[0005] The existing technology has poor real-time performance and the processing speed cannot meet the needs of high-density scenes; in densely populated or complex environments, the existing algorithms are prone to misjudgment or missed judgments; the existing technology focuses more on crowd counting or density estimation, and lacks comprehensive analysis of crowd behavior, flow direction and other dimensions. Therefore, to overcome the above-mentioned defects of the existing technology, the present invention provides an intelligent crowd monitoring system based on deep learning.

[0006] In view of the above problems, the present invention proposes a solution. Summary of the invention

[0007] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an intelligent crowd monitoring system based on deep learning, and solves the problems raised in the above-mentioned background technology through the intelligent crowd monitoring system based on deep learning.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] The intelligent crowd monitoring system based on deep learning includes data acquisition module, video processing module, image recognition module, crowd analysis module, and decision adjustment module. There are connections between the modules:

[0010] The data acquisition module is used to collect monitoring video information in real time, generate monitoring video streams, and transmit the monitoring video streams to the video processing module; the video processing module is used to pre-process the monitoring video streams, use convolutional neural networks (CNN) to perform target detection on the collected monitoring video streams, and extract crowd location and quantity information based on the target detection output results; the image recognition module is used to calculate the crowd density based on the crowd location and quantity information extracted from the video processing module, and track the movement trajectory of the crowd to obtain movement trajectory data, and calculate the crowd distribution change data based on the crowd density;

[0011] The crowd analysis module is used to predict and analyze crowd behavior based on the crowd's movement trajectory and crowd distribution change data using a deep learning model to obtain a predicted congested path; the decision adjustment module is used to calculate the crowd flow speed of the predicted congested path based on the crowd's location information, and to determine whether it is necessary to dynamically adjust the crowd's route based on fuzzy reasoning in combination with the crowd density, and to warn the crowd to avoid congested areas.

[0012] In a preferred embodiment, the method for acquiring the monitoring video stream is as follows: a video acquisition device is installed in each corner of the area; monitoring video information of the area path is acquired in real time to generate an original video signal; the original video signal is encoded and compressed, and converted into a standard video stream format for subsequent transmission and processing to generate a compressed video stream; after the video stream is generated, it is transmitted to a video processing module via a network for further analysis and storage.

[0013] In a preferred embodiment, the steps of using a convolutional neural network (CNN) to detect targets in a collected surveillance video stream are as follows: in a convolution layer, features in the image are extracted through multiple layers of convolution, and the features include edges, corners, and textures; in a pooling layer, the features are reduced in dimension through a pooling operation to reduce the spatial size of the data and improve computational efficiency; the feature maps extracted by the convolution and pooling layers are converted into an output vector of a fixed length, and the output vector includes a bounding box of the crowd, a category label, and a confidence score.

[0014] In a preferred embodiment, the steps of preprocessing the monitoring video stream are as follows: first, extract each frame of image from the monitoring video stream; perform scaling, denoising, and grayscale operations on each frame of image to ensure the quality and consistency of input data.

[0015] In a preferred embodiment, the specific process of extracting crowd location and quantity information based on the target detection output result is as follows: according to the output of the target detection, the crowd location information is extracted through the bounding box in the output vector; for each frame image, when the confidence is greater than a preset threshold, the number of bounding boxes is counted to obtain the number of individuals in the crowd, and the confidence calculation formula is as follows:; wherein, is the number of individuals detected in the t-th frame image, M is the number of all bounding boxes detected in the frame image, is the confidence of the i-th bounding box, and θ is the set confidence threshold.

[0016] In a preferred embodiment, the specific steps of tracking the movement trajectory of the crowd are as follows: based on the crowd position information extracted from the video processing module, the initial position of each individual is obtained, and the next frame position of the target is predicted using Kalman filtering; the position of each individual in consecutive frames is connected, and the trajectory of each target is tracked based on the SORT tracking algorithm, thereby forming movement trajectory data.

[0017] In a preferred embodiment, the process of acquiring the population distribution change data is as follows:

[0018] The entire image is gridded; the density distribution in the area is visualized through a density map; the position of each target is smoothed using a Gaussian kernel to obtain a smooth density distribution; the difference between the density map at two time points is calculated to obtain the population distribution change data.

[0019] In a preferred embodiment, the steps of predicting congested paths are as follows: preprocessing the movement trajectory of the crowd and the crowd distribution change data, removing noise and normalizing, and extracting features useful for prediction from the data; after feature extraction, training the deep learning model through the feature data; after the model training is completed, using the trained model to predict future scenes; predicting the paths that the crowd may pass through in the future and the areas where congestion may occur, and combining the paths that the crowd may pass through and the areas where congestion may occur to obtain a predicted congested path.

[0020] In a preferred embodiment, the steps for obtaining the crowd flow speed are as follows: calculate the moving distance of each individual based on the position information of the crowd, combine the moving distance with the time and the video frame rate to obtain the flow speed of each individual; and obtain the crowd flow speed by weighted averaging the flow speed of each individual.

[0021] In a preferred embodiment, the steps of judging whether it is necessary to dynamically adjust the crowd's route based on fuzzy reasoning are as follows: defining the crowd flow speed and crowd density as input variables, and dividing them into different fuzzy sets respectively; defining whether it is necessary to dynamically adjust the crowd's route as an output variable, and dividing them into fuzzy sets; formulating fuzzy rules to describe the influence of crowd flow speed and crowd density on whether it is necessary to dynamically adjust the crowd's route; performing fuzzy reasoning according to the fuzzy rules to judge whether it is necessary to dynamically adjust the crowd's route.

[0022] Technical effects and advantages of the intelligent crowd monitoring system based on deep learning of the present invention:

[0023] 1. The present invention can capture and analyze the dynamic changes of the crowd in the monitoring area in real time by using deep learning and computer vision technology, and timely identify key information such as the density, flow speed, direction, etc. of the crowd. The system can not only detect the number and distribution of the crowd, but also analyze the behavior patterns of the crowd (such as rapid gathering, slow movement, stay, etc.), and effectively identify abnormal behavior and potential risks. Through image analysis and deep learning, the system can accurately calculate the density of the crowd and identify high-density areas. In the case of too high density, the system can issue a congestion warning to remind relevant personnel to intervene or guide. When it is detected that the density of the flow of people in a specific area exceeds the standard or an abnormal situation occurs, the system can automatically trigger a warning signal, notify the relevant management personnel through sound, image or mobile APP, and take guidance measures in time. The system can intelligently recommend the optimal guidance path or adjust the passage based on real-time monitoring data (such as flow of people, flow speed, etc.) to avoid further aggravation of congested areas. The system can automatically adjust the guidance plan according to the real-time flow of people, improve the efficiency of crowd guidance, and reduce congestion and waiting time.

[0024] In addition to video surveillance data, the system can also integrate other sensor data (such as infrared sensors, thermal imagers, RFID technology, etc.) for multi-source data fusion to improve the detection accuracy and robustness of the system. Based on deep learning image recognition and data fusion algorithms, the system can identify different types of crowd behaviors (such as walking, gathering, queuing, etc.) and respond appropriately.

[0025] 2. The present invention can provide higher accuracy and robustness than traditional methods by extracting deep features from a large amount of data, especially in complex and dynamically changing crowd environments. Based on the adaptive learning ability of neural networks, the system can continuously optimize and improve the accuracy of monitoring and adapt to various changes and different scenarios. The system can automatically identify and analyze the flow status of the crowd without manual intervention, greatly improving the monitoring efficiency and response speed. It can automatically make decisions based on real-time conditions, such as automatically adjusting the diversion path, issuing congestion warnings, etc. The system can monitor 24 hours a day, reducing dependence on manual monitoring, reducing labor costs and the risk of monitoring omissions. The system can predict potential congestion or safety risks in advance, and issue warning signals in time to assist managers to take countermeasures in advance to avoid the occurrence of safety accidents. Through the rapid identification of abnormal behaviors (such as rapid assembly, pushing, running, etc.), the system can capture danger signals in time, prevent and intervene, and improve the safety of public places. The intelligent crowd monitoring system based on deep learning can be applied to various large-scale public places, such as shopping malls, stations, stadiums, concerts, airports, etc., and flexibly respond to different environmental and scene requirements. The system can be customized in venues of different sizes and types, suitable for different monitoring needs, and has strong adaptability and scalability. Through the collection and analysis of real-time data, the system provides managers with accurate decision-making support. For example, understanding the changes in crowd density, formulating more reasonable diversion measures, optimizing crowd distribution, and improving venue operation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a schematic diagram of the structure of the intelligent crowd monitoring system based on deep learning of the present invention. DETAILED DESCRIPTION

[0027] 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.

[0028] Embodiment 1, Figure 1 The present invention provides an intelligent crowd monitoring system based on deep learning.

[0029] The data acquisition module is used to collect monitoring video information in real time, generate monitoring video streams, and transmit the monitoring video streams to the video processing module;

[0030] The video acquisition device is used to collect the monitoring video information of the regional path in real time to generate the original video signal. The above-mentioned video acquisition device is generally a camera (such as an infrared camera, an IP camera (IPC)), which is installed in every corner of the area to achieve the purpose of monitoring all persons entering the venue without blind spots;

[0031] The original video signal needs to be encoded and compressed before it can be converted into a standard video stream format for subsequent transmission and processing;

[0032] The video encoder compresses the original video signal collected in real time according to the selected encoding format to generate a compressed video stream;

[0033] After the video stream is generated, it needs to be transmitted to the video processing module through the network for further analysis, storage or display;

[0034] During video streaming, the following factors usually need to be considered:

[0035] Network bandwidth: High-quality video streaming requires a higher bandwidth. If the bandwidth is insufficient, the picture may be stuck or distorted.

[0036] Latency: Especially in real-time monitoring systems, low-latency transmission is very critical. Signal delays need to be avoided as much as possible during transmission to ensure real-time monitoring information.

[0037] Streaming media server: Video streams may be transferred through streaming media servers to aggregate or distribute surveillance video streams from different sources to multiple processing terminals.

[0038] After the video stream reaches the video processing module, the processing module will perform various operations according to the requirements.

[0039] The video processing module is used to pre-process the surveillance video stream, use the convolutional neural network (CNN) to perform target detection on the collected surveillance video stream, and extract the location and number information of the crowd based on the target detection output results;

[0040] The surveillance video stream needs to be converted into a series of image frames for processing. Usually in video stream processing, videos with higher frame rates will be truncated to reduce the computational burden, and each frame of the image can be used as the input of CNN for analysis.

[0041] The steps of preprocessing the surveillance video stream are as follows:

[0042] First, extract each frame of image from the surveillance video stream;

[0043] Each frame of the image is scaled, denoised, and grayscaled to ensure the quality and consistency of the input data.

[0044] The steps of using convolutional neural network (CNN) to detect the collected surveillance video stream target are as follows:

[0045] In the convolution layer, features in the image are extracted through multi-layer convolution, including edges, corners, and textures. These features are the basis for subsequent determination of the target position.

[0046] In the pooling layer, the feature dimension is reduced through pooling operation, which reduces the spatial size of the data and improves the computational efficiency;

[0047] The feature maps extracted by the convolutional layer and the pooling layer are converted into a fixed-length output vector, which includes the bounding boxes, category labels and confidence scores of the crowd for further classification and regression tasks;

[0048] In crowd detection, the location of the crowd is usually extracted through the bounding box of the target detection, based on which the number of individuals in the crowd can be further calculated.

[0049] According to the output of the target detection network, each target will have a bounding box B, usually expressed as: B = (x min ,y min , x max ,y max ); where (x min ,y min ) is the coordinate of the upper left corner of the bounding box, (x max ,y max ) is the coordinate of the lower right corner of the bounding box, indicating the location of the detected crowd.

[0050] The output of target detection includes the confidence of each detection box (i.e., the probability of classification). The detection box with higher confidence can be selected according to the preset threshold (generally a threshold such as 0.5 is selected); for each frame of image, when the confidence is greater than the preset threshold, the number of bounding boxes is counted to obtain the number of individuals in the crowd. The calculation formula is as follows: Where N t is the number of individuals detected in the t-th frame image, M is the number of all bounding boxes detected in the frame image, and conf i is the confidence of the i-th bounding box, and θ is the set confidence threshold;

[0051] The image recognition module is used to calculate the crowd density based on the crowd position and quantity information extracted from the video processing module, and track the movement trajectory of the crowd to obtain the movement trajectory data, and obtain the crowd distribution change data based on the crowd density calculation;

[0052] SORT is a lightweight online real-time tracking algorithm based on the Hungarian Algorithm to solve the target association problem. It is usually used in combination with Kalman filtering and is suitable for multi-target tracking.

[0053] The specific steps of tracking the movement trajectory of the crowd are as follows:

[0054] According to the crowd position information extracted from the video processing module, the initial position of each individual is obtained, and the Kalman filter is used to predict the next frame position of the target;

[0055] The positions of each individual in consecutive frames are connected, and the trajectory of each target is tracked based on the SORT tracking algorithm to form the movement trajectory data of each individual.

[0056] The entire image is gridded, the crowd density of each grid area is calculated, the density distribution in the area is visualized through the density map, the density of the crowd in each small area is calculated, and the position of each target is smoothed using the Gaussian kernel to obtain a smooth density distribution. The generation formula of the density map is as follows: In the formula, ρ(x, y) represents the density value at (x, y), N is the number of individuals, (x i ,y i ) is the position of the i-th target, σ is the standard deviation of the Gaussian function, which controls the degree of smoothness;

[0057] The difference between the density maps at two time points was calculated to obtain the population distribution change data.

[0058] The crowd analysis module is used to predict and analyze crowd behavior based on the crowd's movement trajectory and crowd distribution change data using a deep learning model to obtain a predicted congestion path;

[0059] Preprocess the movement trajectory and distribution change data of the crowd, remove noise and normalize them, and extract features useful for prediction from the data;

[0060] After feature extraction, the deep learning model is trained with feature data. After model training is completed, the trained model can be used to predict future scenarios.

[0061] Predict the paths that people may take in the future and the areas where congestion may occur, and combine the paths that people may take and the areas where congestion may occur to obtain the predicted congested path.

[0062] The decision adjustment module is used to calculate and predict the crowd flow speed of the congested path based on the crowd's location information, and to determine whether it is necessary to dynamically adjust the crowd's route based on fuzzy reasoning in combination with the crowd density, and to warn the crowd to avoid the congested area.

[0063] The moving distance of each individual can be calculated based on the location information of the crowd. The flow speed of each individual can be obtained by combining the moving distance with the time and the video frame rate. The specific calculation formula for the flow speed of each individual is as follows: In the formula, v i is the flow velocity of each individual, is at time t 1 location, is at time t 2 The position of, f is the video frame rate;

[0064] The flow speed of the crowd is obtained by taking the weighted average of each individual's flow speed;

[0065] The steps to combine the crowd flow speed with the crowd density based on fuzzy reasoning to determine whether the crowd route needs to be dynamically adjusted are as follows:

[0066] In step C1, crowd flow speed and crowd flow density are defined as input variables, and they are divided into different fuzzy sets.

[0067] For example, "Low", "Medium", "High" for crowd flow speed, and "Low", "Medium", "High" for crowd density.

[0068] Step C2, whether it is necessary to dynamically adjust the crowd's route is defined as an output variable, which is divided into fuzzy sets, for example, "No", "Yes", for whether it is necessary to dynamically adjust the crowd's route.

[0069] Step C3, formulate a set of fuzzy rules to describe the impact of different input variables on output variables. The definition of rules can be based on professional knowledge or obtained through data analysis and experiments. For example:

[0070] The crowd flow speed is marked as v, the crowd flow density is marked as ρ, and whether the crowd's route needs to be dynamically adjusted is marked as P.

[0071] Ru le 1:IF(vis Low)AND(ρi s High)THEN(P is Yes)

[0072] Ru le 2:IF(vis High)AND(ρi s Low)THEN(P is No) ...

[0074] Step C4, performing fuzzy reasoning according to fuzzy rules to determine whether it is necessary to dynamically adjust the crowd's route.

[0075] It should be noted that the division of fuzzy sets can be adjusted according to actual conditions. For example, although this embodiment takes three fuzzy sets as examples, in fact, the crowd flow speed, crowd density, and whether the crowd route needs to be dynamically adjusted can be divided into more than three sets to facilitate better and more accurate identification.

[0076] Furthermore, for the judgment of crowd flow speed and crowd density, thresholds can be set according to actual conditions; when the crowd flow speed is higher than 1m / s, it is marked as "High", and when the crowd density is higher than 5 people / square meter, it is marked as "High", and so on. I will not go into details here.

[0077] If it is necessary to warn people to avoid congested areas, a real-time voice broadcast can be issued through mobile devices to remind people to avoid entering congested areas. The real-time situation of congested areas can be displayed through large screens or electronic signs, and alternative routes can be provided.

[0078] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0079] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0080] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0081] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0082] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0083] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. Intelligent crowd monitoring system based on deep learning, characterized by: It includes data acquisition module, video processing module, image recognition module, crowd analysis module, and decision adjustment module. There are connections between the modules: The data acquisition module is used to collect monitoring video information in real time, generate monitoring video streams, and transmit the monitoring video streams to the video processing module; The video processing module is used to pre-process the surveillance video stream, use the convolutional neural network (CNN) to perform target detection on the collected surveillance video stream, and extract the location and number information of the crowd based on the target detection output results; The image recognition module is used to calculate the crowd density based on the crowd position and quantity information extracted from the video processing module, and track the movement trajectory of the crowd to obtain the movement trajectory data, and obtain the crowd distribution change data based on the crowd density calculation; The crowd analysis module is used to predict and analyze crowd behavior based on the crowd's movement trajectory and crowd distribution change data using a deep learning model to obtain a predicted congestion path; The decision adjustment module is used to calculate and predict the crowd flow speed of the congested path based on the crowd's location information, and to determine whether it is necessary to dynamically adjust the crowd's route based on fuzzy reasoning in combination with the crowd density, and to warn the crowd to avoid the congested area.

2. The deep learning-based intelligent crowd monitoring system according to claim 1, characterized in that: The method for obtaining the monitoring video stream is as follows: Install video acquisition devices in every corner of the area; Collect monitoring video information of regional paths in real time to generate original video signals; Encode and compress the original video signal and convert it into a standard video stream format for subsequent transmission and processing to generate a compressed video stream; After the video stream is generated, it is transmitted to the video processing module through the network for further analysis and storage.

3. The deep learning-based intelligent crowd monitoring system according to claim 2 is characterized in that: The steps of using convolutional neural network (CNN) to detect the collected surveillance video stream target are as follows: In the convolution layer, features in the image are extracted through multiple layers of convolution, and the features include edges, corners, and textures; In the pooling layer, the feature dimension is reduced through pooling operation, which reduces the spatial size of the data and improves the computational efficiency; The feature maps extracted by the convolution and pooling layers are converted into a fixed-length output vector, which includes the bounding box of the crowd, the category label, and the confidence score.

4. The deep learning-based intelligent crowd monitoring system according to claim 3 is characterized in that: The steps of preprocessing the surveillance video stream are as follows: First, extract each frame of image from the surveillance video stream; Each frame of the image is scaled, denoised, and grayscaled to ensure the quality and consistency of the input data.

5. The deep learning-based intelligent crowd monitoring system according to claim 4 is characterized in that: The specific process of extracting crowd location and quantity information based on target detection output results is as follows: According to the output of target detection, the crowd location information is extracted through the bounding box in the output vector; For each frame of image, when the confidence is greater than the preset threshold, the number of bounding boxes is counted to obtain the number of individuals in the crowd. The confidence calculation formula is as follows: Where N t is the number of individuals detected in the t-th frame image, M is the number of all bounding boxes detected in the frame image, and conf i is the confidence of the i-th bounding box, and θ is the set confidence threshold.

6. The deep learning-based intelligent crowd monitoring system according to claim 5 is characterized in that: The specific steps of tracking the movement trajectory of the crowd are as follows: According to the crowd position information extracted from the video processing module, the initial position of each individual is obtained, and the Kalman filter is used to predict the next frame position of the target; The positions of each individual in consecutive frames are connected, and the trajectory of each target is tracked based on the SORT tracking algorithm to form the movement trajectory data.

7. The deep learning-based intelligent crowd monitoring system according to claim 6, characterized in that: The process of acquiring the population distribution change data is as follows: Grid the entire image; The density distribution in the area is visualized through the density map; the generation formula of the density map is as follows: In the formula, ρ(x, y) represents the density value at (x, y), N is the number of individuals, (x i ,y i ) is the position of the i-th target, σ is the standard deviation of the Gaussian function, which controls the degree of smoothness; Use a Gaussian kernel to smooth the position of each object to obtain a smooth density distribution; The difference between the density maps at two time points was calculated to obtain the population distribution change data.

8. The deep learning-based intelligent crowd monitoring system according to claim 7, characterized in that: The steps of predicting the congested path are as follows: Preprocess the movement trajectory and distribution change data of the crowd, remove noise and normalize them, and extract features useful for prediction from the data; After feature extraction, the deep learning model is trained using feature data; After the model training is completed, the trained model is used to predict future scenarios; Predict the paths that people may take in the future and the areas where congestion may occur, and combine the paths that people may take and the areas where congestion may occur to obtain the predicted congested path.

9. The deep learning-based intelligent crowd monitoring system according to claim 8, characterized in that: The steps for obtaining the crowd flow speed are as follows: The moving distance of each individual is calculated based on the location information of the crowd, and the flow speed of each individual is obtained by combining the moving distance with the time and the video frame rate; The specific calculation formula for the flow velocity of each individual is as follows: In the formula, v i is the flow velocity of each individual, is at the time t1, is the position at time t2, and f is the video frame rate; The flow speed of the crowd is obtained by taking the weighted average of each individual's flow speed.

10. The deep learning-based intelligent crowd monitoring system according to claim 9, characterized in that: The steps of judging whether it is necessary to dynamically adjust the crowd's route based on fuzzy reasoning are as follows: The crowd flow speed and crowd flow density are defined as input variables, and they are divided into different fuzzy sets respectively; Whether it is necessary to dynamically adjust the crowd's route is defined as the output variable and divided into fuzzy sets; Formulate fuzzy rules to describe the impact of crowd flow speed and crowd density on whether it is necessary to dynamically adjust the crowd's route; Fuzzy reasoning is performed based on fuzzy rules to determine whether the crowd's route needs to be dynamically adjusted.

Images