Scenic spot personnel and vehicle intelligent diversion system and method

Abstract

The invention discloses a scenic spot personnel and vehicle intelligent diversion system and method. The method comprises the following steps: firstly, carrying out key frame discrete sampling on a people flow monitoring video collected by a camera deployed at a scenic spot intersection to obtain a time sequence of people flow monitoring key frames; respectively carrying out feature extraction on the time sequence of the people flow monitoring key frame through a people flow distribution feature extractor based on a deep neural network model so as to obtain a time sequence of a people flow distribution semantic feature graph; secondly, enabling the time sequence of the people flow distribution semantic feature map to pass through a spatial information multi-scale intensive attention reserving module to obtain a time sequence of a spatial saliency people flow distribution semantic feature map; and the time sequence of the spatial saliency people flow distribution semantic feature map passes through a people flow time sequence distribution reasoning module to obtain people flow time sequence reasoning semantic features, and finally, based on the people flow time sequence reasoning semantic features, whether the green light time is prolonged is determined.

Description

Technical Field

[0001] The present disclosure relates to the field of intelligent vehicle diversion, and in particular to a system and method for intelligent diversion of personnel and vehicles in a scenic area. Background Art

[0002] With the rapid development of tourism and the increase in population mobility, scenic spots are crowded with people and vehicles, which seriously affects the efficiency and safety of scenic spots. Effective diversion can allow tourists to enter and exit scenic spots more quickly and conveniently, reduce waiting time, and improve tourists' satisfaction. In addition, reasonable diversion can reduce the incidence of traffic accidents and ensure the safety of tourists and employees. Therefore, the diversion management of people and vehicles in scenic spots has become an important issue for managers.

[0003] However, the traditional way of diverting people and vehicles in scenic spots mainly relies on manual experience and simple rules to make decisions. For example, it relies on manual experience and operation to adjust the green light time of scenic spots. Manual intervention is easily affected by the subjective consciousness and experience of the supervisor, and it is difficult to cope with the complex changes in the flow of people and vehicles, which can easily lead to the green light time being too short or too long. In addition, the traditional way of diverting people and vehicles in scenic spots has a low degree of intelligence, and the response speed of adjusting the green light time is slow. There is often a lag response problem, which cannot respond to emergencies or drastic changes in the density of people flow in a timely manner, which can easily lead to traffic congestion and chaos.

[0004] Therefore, an optimized intelligent diversion system for people and vehicles in scenic spots is desired. Summary of the invention

[0005] The present disclosure is made in consideration of the above problems. One purpose of the present disclosure is to provide a system and method for intelligently guiding people and vehicles in a scenic area.

[0006] The embodiment of the present disclosure provides a smart traffic diversion system for people and vehicles in a scenic spot, which includes:

[0007] A crowd monitoring video acquisition module is used to obtain crowd monitoring videos collected by cameras deployed at intersections in scenic spots;

[0008] A crowd flow monitoring key frame sampling module is used to perform key frame discrete sampling on the crowd flow monitoring video to obtain a time series of crowd flow monitoring key frames;

[0009] A crowd distribution semantic feature extraction module, used for extracting features from each crowd flow monitoring key frame in the time series of the crowd flow monitoring key frames through a crowd flow distribution feature extractor based on a deep neural network model to obtain a time series of crowd flow distribution semantic feature graphs;

[0010] A crowd flow distribution semantic spatial feature attention module is used to obtain a time series of spatially significant crowd flow distribution semantic feature maps by passing the time series of the crowd flow distribution semantic feature maps through a spatial information multi-scale intensive retention attention module;

[0011] A pedestrian flow temporal reasoning module, used for passing the time series of the spatially significant pedestrian flow distribution semantic feature graph through the pedestrian flow temporal distribution reasoning module to obtain a pedestrian flow temporal reasoning semantic feature vector as a pedestrian flow temporal reasoning semantic feature; and

[0012] The green light time control module is used to determine whether to extend the green light time based on the semantic features of the pedestrian flow temporal reasoning.

[0013] For example, according to the intelligent guidance system for people and vehicles in a scenic area according to an embodiment of the present disclosure, the deep neural network model is a convolutional neural network model.

[0014] For example, according to the intelligent traffic diversion system for scenic area personnel and vehicles in an embodiment of the present disclosure, the crowd distribution semantic space feature attention module includes:

[0015] A crowd distribution semantic feature map multi-mode pooling unit, used for performing multi-mode pooling processing on each crowd distribution semantic feature map in the time series of the crowd distribution semantic feature map along the channel dimension to obtain a spliced ​​crowd distribution semantic feature matrix;

[0016] A crowd flow distribution implicit correlation feature capturing unit, used for performing convolution processing on the spliced ​​crowd flow distribution semantic feature matrix through a convolution layer to obtain a crowd flow distribution implicit correlation feature matrix;

[0017] A crowd flow distribution spatial attention weight calculation activation unit, used for inputting the crowd flow distribution implicit correlation feature matrix into a Sigmoid activation unit to obtain a crowd flow distribution spatial attention weight matrix; and

[0018] The spatial saliency enhancement unit is used to multiply each feature matrix along the channel dimension in the crowd distribution semantic feature map by the crowd distribution spatial attention weight matrix according to the position point to obtain the spatially salient crowd distribution semantic feature map.

[0019] For example, according to the intelligent flow guidance system for scenic spot personnel and vehicles in an embodiment of the present disclosure, the multi-mode pooling unit of the semantic feature graph of the flow distribution includes:

[0020] A global mean pooling subunit, used for performing global mean pooling processing on the crowd flow distribution semantic feature map along the channel dimension to obtain a crowd flow distribution semantic global mean pooling feature matrix;

[0021] A maximum pooling subunit, used for performing maximum pooling processing on the crowd flow distribution semantic feature map along the channel dimension to obtain a crowd flow distribution semantic maximum pooling feature matrix;

[0022] A random pooling subunit, configured to perform random pooling processing on the crowd flow distribution semantic feature map along the channel dimension to obtain a crowd flow distribution semantic random pooling feature matrix; and

[0023] The multi-pooling semantic splicing unit is used to splice the crowd distribution semantic mean pooling feature matrix, the crowd distribution semantic maximum pooling feature matrix and the crowd distribution semantic random pooling feature matrix to obtain the spliced ​​crowd distribution semantic feature matrix.

[0024] For example, according to the intelligent traffic diversion system for scenic spot personnel and vehicles in an embodiment of the present disclosure, the pedestrian flow time sequence reasoning module includes:

[0025] A vectorization expansion unit, used for vectorizing and expanding a plurality of spatially significant crowd flow distribution semantic feature graphs in the time series of the spatially significant crowd flow distribution semantic feature graphs to obtain a sequence of spatially significant crowd flow distribution semantic feature vectors;

[0026] A vector screening unit, used to determine the spatially significant crowd flow distribution semantic feature vector at the last position in the sequence of the spatially significant crowd flow distribution semantic feature vectors;

[0027] A crowd flow distribution average information difference calculation unit is used to calculate the position difference between the mean feature vector of multiple spatially significant crowd flow distribution semantic feature vectors in the sequence of the spatially significant crowd flow distribution semantic feature vectors and the spatially significant crowd flow distribution semantic feature vector at the last position to obtain a crowd flow distribution average information difference feature vector;

[0028] A random information difference calculation unit for crowd flow distribution is used to calculate the position difference between the spatially significant crowd flow distribution semantic feature vector at a random position in the sequence of the spatially significant crowd flow distribution semantic feature vector and the spatially significant crowd flow distribution semantic feature vector at the last position to obtain a random information difference feature vector for crowd flow distribution;

[0029] A first crowd flow distribution inference unit, configured to multiply the crowd flow distribution average information difference feature vector by a first predetermined weight hyperparameter and a first Gaussian distribution random number function value to obtain a first crowd flow distribution inference feature vector;

[0030] A second crowd flow distribution inference unit, configured to multiply the crowd flow distribution random information difference feature vector by a second predetermined weight hyperparameter and a second Gaussian distribution random number function value to obtain a second crowd flow distribution inference feature vector; and

[0031] A fusion unit is used to calculate the positional addition between the first crowd flow distribution reasoning feature vector and the second crowd flow distribution reasoning feature vector to obtain the crowd flow temporal reasoning semantic feature vector.

[0032] For example, according to the intelligent guidance system for scenic area personnel and vehicles according to the embodiment of the present disclosure, the first Gaussian distribution random number function value and the second Gaussian distribution random number function value are generated by a Gaussian distribution random number function with a mean of 0 and a variance of 1.

[0033] For example, according to the intelligent traffic diversion system for scenic spot personnel and vehicles in an embodiment of the present disclosure, the green light time control module is used to:

[0034] The pedestrian flow time series reasoning semantic feature vector is passed through a traffic light controller based on a classifier to obtain a control instruction, where the control instruction is used to indicate whether to extend the green light time.

[0035] For example, according to the intelligent traffic guidance system for scenic area personnel and vehicles of the embodiment of the present disclosure, it also includes a training module for training the crowd distribution feature extractor based on the convolutional neural network model, the spatial information multi-scale intensive retention attention module, the crowd flow time series distribution reasoning module and the classifier-based traffic light controller.

[0036] For example, according to the intelligent guidance system for scenic spot personnel and vehicles in an embodiment of the present disclosure, the training module includes:

[0037] A training data acquisition unit, used to acquire training data, wherein the training data includes a training flow monitoring video collected by a camera deployed at a scenic spot intersection;

[0038] A training crowd flow monitoring key frame sampling unit, used for performing key frame discrete sampling on the training crowd flow monitoring video to obtain a time series of training crowd flow monitoring key frames;

[0039] A training crowd flow distribution semantic feature extraction unit is used to extract features from each training crowd flow monitoring key frame in the time series of the training crowd flow monitoring key frames through the crowd flow distribution feature extractor based on the convolutional neural network model to obtain a time series of training crowd flow distribution semantic feature graphs;

[0040] A training crowd flow distribution semantic space feature attention unit is used to pass the time series of the training crowd flow distribution semantic feature map through the spatial information multi-scale dense retention attention module to obtain a time series of training space-significant crowd flow distribution semantic feature map;

[0041] A training pedestrian flow time series reasoning unit is used to pass the time series of the training spatially significant pedestrian flow distribution semantic feature graph through the pedestrian flow time series distribution reasoning module to obtain a training pedestrian flow time series reasoning semantic feature vector;

[0042] A loss calculation unit, used for passing the training pedestrian flow temporal reasoning semantic feature vector through the classifier-based traffic light controller to obtain a classification loss function value; and

[0043] A loss training unit is used to train the crowd distribution feature extractor based on the convolutional neural network model, the spatial information multi-scale dense retention attention module, the crowd flow temporal distribution reasoning module and the classifier-based traffic light controller based on the classification loss function value, wherein in each round of iteration of the training, the training crowd flow temporal reasoning semantic feature vector is corrected.

[0044] The embodiment of the present disclosure also provides a method for intelligently directing people and vehicles in a scenic spot, which includes:

[0045] Obtain the crowd monitoring video collected by the cameras deployed at the intersections of the scenic area;

[0046] Performing key frame discrete sampling on the crowd flow monitoring video to obtain a time series of crowd flow monitoring key frames;

[0047] By using a crowd distribution feature extractor based on a deep neural network model, feature extraction is performed on each crowd flow monitoring key frame in the time series of the crowd flow monitoring key frames to obtain a time series of crowd flow distribution semantic feature graphs;

[0048] The time series of the semantic feature map of crowd flow distribution is passed through a spatial information multi-scale dense retention attention module to obtain a time series of spatially significant semantic feature maps of crowd flow distribution;

[0049] Passing the time series of the spatially significant crowd flow distribution semantic feature graph through a crowd flow temporal distribution reasoning module to obtain a crowd flow temporal reasoning semantic feature vector as a crowd flow temporal reasoning semantic feature; and

[0050] Based on the pedestrian flow time series reasoning semantic features, determine whether to extend the green light time.

[0051] According to the intelligent personnel and vehicle diversion system and method for scenic spots in the embodiments of the present disclosure, the system collects crowd flow monitoring videos in real time through cameras arranged at intersections of scenic spots, and introduces image processing and analysis algorithms based on artificial intelligence and machine vision at the back end to analyze the crowd flow monitoring videos, so as to capture semantic features related to crowd flow distribution and density. Then, the crowd flow distribution semantics extracted from the crowd flow monitoring videos are enhanced based on multi-scale correlation features in the time-space dimension, so as to infer the time series pattern of crowd flow distribution and predict the next crowd flow situation, so as to be able to intelligently adjust the green light time, improve the operation efficiency of the scenic spot and the tourist experience, and reduce the probability of accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following briefly introduces the drawings of the embodiments of the present disclosure. Obviously, the drawings described below only relate to some embodiments of the present disclosure, and are not intended to limit the present disclosure.

[0053] Figure 1 A schematic diagram of the application architecture of the intelligent traffic diversion system for scenic area personnel and vehicles in the embodiment of the present disclosure is shown;

[0054] Figure 2 A schematic diagram of the structure of the intelligent diversion system for people and vehicles in a scenic area according to an embodiment of the present disclosure is shown;

[0055] Figure 3 A schematic diagram showing the structure of a semantic space feature attention module for the flow distribution of people in a scenic area and a smart flow diversion system for people and vehicles in an embodiment of the present disclosure is shown;

[0056] Figure 4 A schematic diagram of the structure of a pedestrian flow time sequence reasoning module of a scenic spot personnel and vehicle intelligent flow diversion system in an embodiment of the present disclosure is shown;

[0057] Figure 5 A flow chart of a method for intelligently directing people and vehicles in a scenic area according to an embodiment of the present disclosure is shown;

[0058] Figure 6 An application scenario diagram of the intelligent traffic diversion system for scenic area personnel and vehicles in the embodiment of the present disclosure is shown; and

[0059] Figure 7 A schematic diagram of a storage medium according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present disclosure will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present disclosure, rather than all embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present disclosure.

[0061] The terms used in this specification are those general terms currently widely used in the art in consideration of the functions of the present disclosure, but these terms may vary according to the intention of a person of ordinary skill in the art, precedents, or new technologies in the art. In addition, specific terms may be selected by the applicant, and in this case, their detailed meanings will be described in the detailed description of the present disclosure. Therefore, the terms used in the specification should not be understood as simple names, but rather as an overall description based on the meaning of the terms and the present disclosure.

[0062] Although the present disclosure makes various references to certain modules in the system according to embodiments of the present disclosure, any number of different modules can be used and run on a user terminal and / or server. The modules are illustrative only, and different aspects of the systems and methods can use different modules.

[0063] Flowcharts are used in this disclosure to illustrate the operations performed by the system according to the embodiments of the present disclosure. It should be understood that the preceding or following operations are not necessarily performed precisely in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. At the same time, other operations may also be added to these processes, or one or more operations may be removed from these processes.

[0064] Figure 1 A schematic diagram of the application architecture of the intelligent guidance system for scenic area personnel and vehicles in an embodiment of the present disclosure is shown, including a server 100 and a terminal device 200.

[0065] The terminal device 200 and the server 100 can be connected via the Internet to achieve mutual communication. Optionally, the above-mentioned Internet uses standard communication technology and / or protocol. The Internet is usually the Internet, but it can also be any network, including but not limited to any combination of a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a dedicated network or a virtual private network. In some embodiments, the data exchanged through the network is represented by technologies and / or formats including HyperText Markup Language (HTML), Extensible Markup Language (XML), etc. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPsec) can also be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can also be used to replace or supplement the above-mentioned data communication technologies.

[0066] The server 100 can provide various network services for the terminal device 200, wherein the server 100 can be a single server, a server cluster consisting of several servers, or a cloud computing center. Specifically, the server 100 may include a processor 110 (Center Processing Unit, CPU), a memory 120, an input device 130, and an output device 140, etc. The input device 130 may include a keyboard, a mouse, a touch screen, etc., and the output device 140 may include a display device, such as a liquid crystal display (Liquid Crystal Display, LCD), a cathode ray tube (Cathode Ray Tube, CRT), etc.

[0067] The memory 120 may include a read-only memory (ROM) and a random access memory (RAM), and provides the program instructions and data stored in the memory 120 to the processor 110. In the embodiment of the present disclosure, the memory 120 may be used to store the corresponding method of the intelligent diversion system for scenic spot personnel and vehicles in the embodiment of the present disclosure.

[0068] The processor 110 calls the program instructions stored in the memory 120, and the processor 110 is used to execute the steps of any one of the methods for intelligently directing people and vehicles in a scenic area in accordance with the obtained program instructions in the embodiments of the present disclosure.

[0069] In addition, the application architecture diagram in the embodiment of the present disclosure is intended to more clearly illustrate the technical solution in the embodiment of the present disclosure, and does not constitute a limitation on the technical solution provided in the embodiment of the present disclosure. Of course, for other application architectures and business applications, the technical solution provided in the embodiment of the present disclosure is also applicable to similar problems.

[0070] The following is a non-restrictive description of the intelligent guidance system for people and vehicles in a scenic area provided according to at least one embodiment of the present disclosure through several examples or embodiments. As described below, different features in these specific examples or embodiments can be combined with each other without conflicting with each other to obtain new examples or embodiments, and these new examples or embodiments also fall within the scope of protection of the present disclosure.

[0071] In response to the above technical problems, in the technical solution of this application, a smart flow diversion system for scenic area personnel and vehicles is proposed, which can collect crowd flow monitoring videos in real time through cameras arranged at intersections of scenic areas, and introduce image processing and analysis algorithms based on artificial intelligence and machine vision at the back end to analyze the crowd flow monitoring videos, so as to capture semantic features related to crowd flow distribution and density. Then, by enhancing the crowd flow distribution semantics extracted from the crowd flow monitoring videos based on multi-scale correlation features in the time-space dimension, the time series pattern of crowd flow distribution is inferred, and the next crowd flow situation is predicted, so as to be able to intelligently adjust the green light time, improve the operation efficiency of the scenic area and the tourist experience, and reduce the probability of accidents.

[0072] Figure 2The structure diagram of the intelligent flow diversion system 800 for scenic area personnel and vehicles in the embodiment of the present disclosure is shown. The intelligent flow diversion system 800 for scenic area personnel and vehicles includes: a crowd flow monitoring video acquisition module 810, which is used to obtain the crowd flow monitoring video collected by the camera deployed at the intersection of the scenic area; a crowd flow monitoring key frame sampling module 820, which is used to perform key frame discrete sampling on the crowd flow monitoring video to obtain a time series of crowd flow monitoring key frames; a crowd flow distribution semantic feature extraction module 830, which is used to extract features from each crowd flow monitoring key frame in the time series of the crowd flow monitoring key frames through a crowd flow distribution feature extractor based on a deep neural network model to obtain a crowd flow distribution semantic feature map. time series; a crowd distribution semantic spatial feature attention module 840, which is used to pass the time series of the crowd distribution semantic feature map through the spatial information multi-scale intensive retention attention module to obtain the time series of the spatially significant crowd distribution semantic feature map; a crowd temporal reasoning module 850, which is used to pass the time series of the spatially significant crowd distribution semantic feature map through the crowd flow temporal distribution reasoning module to obtain the crowd flow temporal reasoning semantic feature vector as the crowd flow temporal reasoning semantic feature; and a green light time control module 860, which is used to determine whether to extend the green light time based on the crowd flow temporal reasoning semantic feature.

[0073] Specifically, in the technical solution of the present application, first, a crowd flow monitoring video collected by a camera deployed at a scenic spot intersection is obtained. Next, considering that the crowd flow monitoring video is usually a continuous video stream, it contains a large amount of redundant information and interference information. Therefore, it is necessary to perform key frame discrete sampling on the crowd flow monitoring video to obtain a time series of crowd flow monitoring key frames. In particular, here, the key frame refers to a representative frame image in the video, which contains important crowd flow distribution information in crowd flow monitoring, such as personnel density, flow direction, surge in crowd flow and periodic change nodes. Through key frame discrete sampling, the amount of data processed for video data can be effectively reduced, the data processing efficiency can be improved, and these key crowd flow distribution information can be better retained and analyzed, which is beneficial to the diversion management of personnel and vehicles in the scenic area.

[0074] Then, considering that each of the crowd flow monitoring key frames in the time series of the crowd flow monitoring key frames contains semantic information about crowd flow distribution, such as crowd density, flow direction of crowd flow, etc., this helps to further understand the semantics and temporal change rules of crowd flow distribution in the future. Based on this, in the technical solution of the present application, each of the crowd flow monitoring key frames in the time series of the crowd flow monitoring key frames is subjected to feature mining in a crowd flow distribution feature extractor based on a convolutional neural network model, so as to extract the semantic feature information of crowd flow distribution in each of the crowd flow monitoring key frames, thereby obtaining a time series of crowd flow distribution semantic feature graphs. The crowd flow distribution semantic feature graph contains semantic features about crowd flow distribution under specific key frames, which can provide richer information support for the subsequent capture of the temporal pattern and change trend of crowd flow distribution, and is conducive to more accurate green light time adjustment and crowd flow guidance.

[0075] Correspondingly, in the crowd distribution semantic feature extraction module 830, the deep neural network model is a convolutional neural network model, that is, the crowd distribution feature extractor based on the deep neural network model is a crowd distribution feature extractor based on the convolutional neural network model.

[0076] Furthermore, considering that in the actual process of crowd flow monitoring and crowd flow distribution reasoning, it is necessary to focus more on the spatial position characteristics of the crowd flow distribution in each key frame, such as information such as densely populated areas and crowd flow paths. Based on this, in the technical solution of the present application, the time series of the crowd flow distribution semantic feature map is further passed through the spatial information multi-scale intensive retention attention module to obtain the time series of spatially significant crowd flow distribution semantic feature map. It should be understood that through the processing of the spatial information multi-scale intensive retention attention module, the model can better pay attention to the spatially significant information in the crowd flow distribution key frame, that is, highlight and emphasize the areas of great significance such as densely populated areas and crowd flow paths. At the same time, the multi-scale intensive retention attention module can also capture the spatial feature information at different scales in the key frame, and aggregate and integrate this information, so that the model can understand the semantics of crowd flow distribution more comprehensively and improve the representation ability of complex scenes. In this way, it can provide a more spatially significant crowd flow distribution semantic feature representation for subsequent decision-making, help to analyze the spatial distribution characteristics of crowd flow more finely, and provide more powerful support for intelligent decision-making.

[0077] Accordingly, if Figure 3As shown, the crowd distribution semantic space feature attention module 840 includes: a crowd distribution semantic feature map multi-mode pooling unit 841, which is used to perform multi-mode pooling processing on each crowd distribution semantic feature map in the time series of the crowd distribution semantic feature map along the channel dimension to obtain a spliced ​​crowd distribution semantic feature matrix; a crowd distribution implicit correlation feature capture unit 842, which is used to perform convolution processing on the spliced ​​crowd distribution semantic feature matrix through a convolution layer to obtain a crowd distribution implicit correlation feature matrix; a crowd distribution spatial attention weight calculation activation unit 843, which is used to input the crowd distribution implicit correlation feature matrix into a Sigmoid activation unit to obtain a crowd distribution spatial attention weight matrix; and a spatial saliency enhancement unit 844, which is used to perform position point multiplication of each feature matrix along the channel dimension in the crowd distribution semantic feature map with the crowd distribution spatial attention weight matrix to obtain the spatially salient crowd distribution semantic feature map.

[0078] Among them, the multi-mode pooling unit 841 of the crowd distribution semantic feature map includes: a global mean pooling subunit, which is used to perform global mean pooling processing on the crowd distribution semantic feature map along the channel dimension to obtain a crowd distribution semantic global mean pooling feature matrix; a maximum pooling subunit, which is used to perform maximum pooling processing on the crowd distribution semantic feature map along the channel dimension to obtain a crowd distribution semantic maximum pooling feature matrix; a random pooling subunit, which is used to perform random pooling processing on the crowd distribution semantic feature map along the channel dimension to obtain a crowd distribution semantic random pooling feature matrix; and a multi-pooling semantic splicing unit, which is used to splice the crowd distribution semantic mean pooling feature matrix, the crowd distribution semantic maximum pooling feature matrix and the crowd distribution semantic random pooling feature matrix to obtain the spliced ​​crowd distribution semantic feature matrix.

[0079] Specifically, the crowd distribution semantic spatial feature attention module 840 is used to: process the time series of the crowd distribution semantic feature map through the spatial information multi-scale dense retention attention module with the following spatial explicitation formula to obtain the time series of the spatially significant crowd distribution semantic feature map; wherein the spatial explicitation formula is:

[0080]

[0081] Among them, F2 is each crowd flow distribution semantic feature graph in the time series of the crowd flow distribution semantic feature graph, f N ×N represents the convolution kernel size, Respectively represent the mean pooling feature matrix, the maximum pooling feature matrix and the random pooling feature matrix obtained by performing global mean pooling, global maximum pooling and global random pooling on each of the semantic feature maps of the crowd distribution along the channel dimension, σ is the Sigmoid activation function, M s (F2) is the attention weight matrix for each of the semantic feature maps of the crowd distribution, represents the Hadamard product, and F2′ is each spatially significant crowd flow distribution semantic feature map in the time series of the spatially significant crowd flow distribution semantic feature map.

[0082] It should be understood that since the change of human flow has obvious temporal characteristics, that is, there is a temporal correlation relationship between the semantic features of human flow distribution under each key frame after spatial saliency enhancement. Therefore, in order to better characterize the temporal distribution characteristics of the semantics of spatial distribution of human flow, including the semantic information on the time dimension such as temporal fluctuation, trend and periodicity of human flow, and provide richer feature representation for subsequent analysis and decision-making, in the technical solution of the present application, the time series of the spatially significant human flow distribution semantic feature map is further passed through the human flow temporal distribution reasoning module to obtain the human flow temporal reasoning semantic feature vector. Through the processing of the human flow temporal distribution reasoning module, the semantic features of human flow distribution of each key frame after spatial saliency enhancement can be temporally associated and understood, so as to better understand the dynamic change law of human flow distribution semantics, and use the temporal correlation features of the spatial distribution of human flow as prior information to perform human flow reasoning prediction in the next short period of time, so as to provide richer information support for subsequent decision-making, such as intelligently adjusting the green light time, optimizing human flow guidance, etc., and improving the intelligent level of scenic spot management.

[0083] Accordingly, if Figure 4As shown, the crowd flow time series reasoning module 850 includes: a vector expansion unit 851, which is used to vectorize and expand multiple spatially significant crowd flow distribution semantic feature maps in the time series of the spatially significant crowd flow distribution semantic feature map to obtain a sequence of spatially significant crowd flow distribution semantic feature vectors; a vector screening unit 852, which is used to determine the spatially significant crowd flow distribution semantic feature vector of the last position in the sequence of the spatially significant crowd flow distribution semantic feature vectors; a crowd flow distribution average information difference calculation unit 853, which is used to calculate the position difference between the mean feature vector of multiple spatially significant crowd flow distribution semantic feature vectors in the sequence of the spatially significant crowd flow distribution semantic feature vectors and the spatially significant crowd flow distribution semantic feature vector of the last position to obtain the crowd flow distribution average information difference feature vector; a crowd flow distribution random information difference calculation unit 854, which is used to calculate the spatial The spatially significant crowd flow distribution semantic feature vector at a random position in the sequence of the significant crowd flow distribution semantic feature vector is used to obtain the random information difference feature vector of crowd flow distribution by the position difference between the spatially significant crowd flow distribution semantic feature vector at the last position; the first crowd flow distribution reasoning unit 855 is used to multiply the crowd flow distribution average information difference feature vector by the first predetermined weight hyperparameter and the first Gaussian distribution random number function value to obtain the first crowd flow distribution reasoning feature vector; the second crowd flow distribution reasoning unit 856 is used to multiply the crowd flow distribution random information difference feature vector by the second predetermined weight hyperparameter and the second Gaussian distribution random number function value to obtain the second crowd flow distribution reasoning feature vector; and the fusion unit 857 is used to calculate the positional addition between the first crowd flow distribution reasoning feature vector and the second crowd flow distribution reasoning feature vector to obtain the crowd flow time series reasoning semantic feature vector. Wherein, the first Gaussian distribution random number function value and the second Gaussian distribution random number function value are generated by a Gaussian distribution random number function with a mean of 0 and a variance of 1.

[0084] Specifically, the pedestrian flow time series reasoning module 850 is used to: process the time series of the spatially significant pedestrian flow distribution semantic feature map through the pedestrian flow time series distribution reasoning module using the following reasoning formula to obtain the pedestrian flow time series reasoning semantic feature vector; wherein the reasoning formula is:

[0085]

[0086] Wherein, V1 is the spatially significant crowd flow distribution semantic feature vector at the last position of the sequence of spatially significant crowd flow distribution semantic feature vectors obtained by vectorizing and expanding multiple spatially significant crowd flow distribution semantic feature maps in the time series of the spatially significant crowd flow distribution semantic feature map, V randis the spatially significant crowd flow distribution semantic feature vector at a random position in the sequence of the spatially significant crowd flow distribution semantic feature vectors, V r is the mean feature vector of the sequence of the spatially significant crowd flow distribution semantic feature vectors, p1 and p2 are weight hyperparameters, f1 and f2 are hyperparameters that use a Gaussian distribution random number function with a mean of 0 and a variance of 1 as the coefficients of the Gaussian distribution function, and are vector subtraction and vector addition, V i+1 A semantic feature vector for temporal reasoning of the pedestrian flow.

[0087] Then, the semantic feature vector of the temporal reasoning of the flow of people is passed through the traffic light controller based on the classifier to obtain a control instruction, which is used to indicate whether to extend the green light time. In other words, the temporal reasoning semantic features of the flow of people are used for classification processing to determine whether to extend the green light time. By intelligently adjusting the green light time, the operating efficiency of the scenic spot and the tourist experience can be improved, and the probability of accidents can be reduced.

[0088] Correspondingly, the green light time control module 860 is used to: pass the pedestrian flow time series reasoning semantic feature vector through a classifier-based traffic light controller to obtain a control instruction, and the control instruction is used to indicate whether to extend the green light time.

[0089] Specifically, the green light time control module 860 is used to: use the fully connected layer of the classifier-based traffic light controller to fully connect the pedestrian flow temporal reasoning semantic feature vector to obtain an encoded classification feature vector; and input the encoded classification feature vector into the Softmax classification function of the classifier-based traffic light controller to obtain the control instruction.

[0090] It should be understood that the role of the classifier is to use the given categories and known training data to learn classification rules and classifiers, and then classify (or predict) unknown data. Logistic regression and SVM are often used to solve binary classification problems. For multi-class classification problems, logistic regression or SVM can also be used, but multiple binary classifications are required to form a multi-classification. However, this is prone to errors and is not efficient. Commonly used multi-classification methods include the Softmax classification function.

[0091] Furthermore, in the technical solution disclosed in the present invention, the intelligent traffic diversion system for scenic area personnel and vehicles also includes a training module for training the crowd distribution feature extractor based on the convolutional neural network model, the spatial information multi-scale intensive retention attention module, the crowd flow time series distribution reasoning module and the classifier-based traffic light controller.

[0092] The training module includes: a training data acquisition unit for acquiring training data, wherein the training data includes a training crowd flow monitoring video collected by a camera deployed at a scenic spot intersection; a training crowd flow monitoring key frame sampling unit for performing key frame discrete sampling on the training crowd flow monitoring video to obtain a time series of training crowd flow monitoring key frames; a training crowd flow distribution semantic feature extraction unit for performing feature extraction on each training crowd flow monitoring key frame in the time series of the training crowd flow monitoring key frames through the crowd flow distribution feature extractor based on the convolutional neural network model to obtain a time series of training crowd flow distribution semantic feature maps; a training crowd flow distribution semantic spatial feature attention unit for performing feature extraction on the time series of the training crowd flow distribution semantic feature maps through the spatial information multi-scale dense retention attention module to obtain a training space A time series of spatially significant semantic feature maps of pedestrian flow distribution; a training pedestrian flow temporal reasoning unit, used to pass the time series of the training spatially significant semantic feature maps of pedestrian flow distribution through the pedestrian flow temporal distribution reasoning module to obtain a training pedestrian flow temporal reasoning semantic feature vector; a loss calculation unit, used to pass the training pedestrian flow temporal reasoning semantic feature vector through the classifier-based traffic light controller to obtain a classification loss function value; and a loss training unit, used to train the pedestrian flow distribution feature extractor based on the convolutional neural network model, the spatial information multi-scale densely retained attention module, the pedestrian flow temporal distribution reasoning module and the classifier-based traffic light controller based on the classification loss function value, wherein in each round of iteration of the training, the training pedestrian flow temporal reasoning semantic feature vector is corrected.

[0093] In the above technical scheme, the time series of the training crowd flow distribution semantic feature map represents the image semantic features of each training crowd flow monitoring key frame, and after the time series of the training crowd flow distribution semantic feature map passes through the spatial information multi-scale dense retention attention module and the crowd flow temporal distribution reasoning module, the temporal distribution pattern reasoning between feature matrices can be performed based on the image semantic spatial distribution multi-scale dense attention enhancement of the image semantic features of the local training crowd flow monitoring key frame in the feature matrix. Therefore, the training crowd flow temporal reasoning semantic feature vector will have the complexity of expressing the spatiotemporal cross-dimensional correlation features of the multi-scale correlation features under the temporal-spatial dimension, that is, the training crowd flow temporal reasoning semantic feature vector has rich geometric shape details in the high-dimensional feature space, so that there is category representation distortion when it is classified by the classifier to map to the class probability domain, that is, it affects the category representation of the training crowd flow temporal reasoning semantic feature vector for the category predetermined by the classifier, thereby reducing the accuracy of the classification result.

[0094] Based on this, the applicant of the present application corrects the training pedestrian flow temporal reasoning semantic feature vector each time the training pedestrian flow temporal reasoning semantic feature vector is classified and iterated through a classifier-based traffic light controller.

[0095] Accordingly, in one example, in each iteration of the training, the training pedestrian flow temporal reasoning semantic feature vector is corrected by the following correction formula to obtain a corrected training pedestrian flow temporal reasoning semantic feature vector; wherein the correction formula is:

[0096]

[0097] Wherein, V is the semantic feature vector of the training pedestrian flow temporal reasoning, p is the class probability value obtained by the traffic light controller based on the classifier for the semantic feature vector V of the training pedestrian flow temporal reasoning, and v i is the i-th eigenvalue of the training pedestrian flow temporal reasoning semantic feature vector V, is the mean of all feature values ​​of the semantic feature vector V for temporal reasoning of the training flow, ‖V‖1 represents the norm of the semantic feature vector V for temporal reasoning of the training flow, and α is a weight hyperparameter, exp(·) represents the exponential operation of a value, and the exponential operation of the value represents the calculation of the natural exponential function value with the value as the power, v i ′ It is the i-th eigenvalue of the corrected training pedestrian flow temporal reasoning semantic feature vector.

[0098] Specifically, while maintaining the geometric shape details of the high-dimensional feature manifold under the class probability mapping of the training pedestrian flow time-series reasoning semantic feature vector V, the shape attribute editing is performed on the feature value details relative to the variational representation of the feature set of the training pedestrian flow time-series reasoning semantic feature vector V by means of class probability distortion consultation, thereby mapping the class probability distortion of the training pedestrian flow time-series reasoning semantic feature vector V to the latent class space feature representation, and then consulting fusion is performed by supplementing the basic low-rank constrained latent inversion representation to bridge the gap between the edited consulting representation and the original geometric shape details, thereby improving the category representation of the high-dimensional features of the training pedestrian flow time-series reasoning semantic feature vector V, thereby improving the accuracy of the classification results. In this way, the temporal pattern of pedestrian flow distribution can be more accurately inferred based on the temporal-spatial dimensional characteristics of pedestrian flow distribution, and the next pedestrian flow situation can be predicted, so as to be able to intelligently adjust the green light time, improve the operating efficiency of the scenic spot and the tourist experience, and reduce the probability of accidents.

[0099] Based on the above embodiments, see Figure 5 As shown in FIG. 1 , a flow chart of a method for intelligently directing people and vehicles in a scenic spot according to an embodiment of the present disclosure is shown. For example, the method for intelligently directing people and vehicles in a scenic spot can be executed by a server, which can be Figure 1 The server 100 shown in FIG. Figure 5 As shown, according to the embodiment of the present disclosure, the method for intelligently directing personnel and vehicles in a scenic area includes the following steps: S510, obtaining a crowd monitoring video collected by a camera deployed at a scenic intersection; S520, performing discrete sampling of key frames on the crowd monitoring video to obtain a time series of crowd monitoring key frames; S530, performing feature extraction on each crowd monitoring key frame in the time series of the crowd monitoring key frames through a crowd distribution feature extractor based on a deep neural network model to obtain a time series of crowd distribution semantic feature maps; S540, passing the time series of the crowd distribution semantic feature maps through a spatial information multi-scale dense retention attention module to obtain a time series of spatially significant crowd distribution semantic feature maps; S550, passing the time series of the spatially significant crowd distribution semantic feature maps through a crowd temporal distribution inference module to obtain a crowd temporal inference semantic feature vector as a crowd temporal inference semantic feature; and, S560, determining whether to extend the green light time based on the crowd temporal inference semantic feature.

[0100] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned scenic spot personnel and vehicle intelligent diversion method have been referenced above. Figures 2 to 4 It has been introduced in detail in the description of the scenic area personnel and vehicle intelligent guidance system 800, and therefore, its repeated description will be omitted.

[0101] Figure 6 FIG. 1 is an application scenario diagram of the intelligent traffic diversion system for scenic area personnel and vehicles according to an embodiment of the present disclosure. Figure 6 As shown, in this application scenario, first, a crowd flow monitoring video collected by a camera deployed at a scenic spot intersection is obtained (for example, Figure 6 D), then, the crowd flow monitoring video is input to a server (for example, Figure 6 In S) shown in , the server can use the scenic area personnel and vehicle intelligent diversion algorithm to process the crowd flow monitoring video to obtain a control instruction for indicating whether to extend the green light time.

[0102] Based on the above embodiments, another exemplary embodiment of an electronic device is also provided in the embodiments of the present disclosure. In some possible implementations, the electronic device in the embodiments of the present disclosure may include a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor can implement the steps of the method for intelligently directing people and vehicles in the scenic area in the above embodiments when executing the program.

[0103] For example, the electronic device is disclosed in the present invention Figure 1 Taking the server 100 in the example as an example, the processor in the electronic device is the processor 110 in the server 100, and the memory in the electronic device is the memory 120 in the server 100.

[0104] An embodiment of the present disclosure also provides a computer-readable storage medium. Figure 7 FIG. 1 is a schematic diagram of a computer-readable storage medium 1000 according to an embodiment of the present disclosure. Figure 7 As shown, the computer-readable storage medium 1000 stores computer-executable instructions 1001. When the computer-executable instructions 1001 are executed by the processor, the method for intelligent diversion of personnel and vehicles in the scenic area according to the embodiment of the present disclosure described with reference to the above figures can be executed. The computer-readable storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. The volatile memory may, for example, include a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may, for example, include a read-only memory (ROM), a hard disk, a flash memory, etc.

[0105] The embodiments of the present disclosure also provide a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method for intelligently guiding people and vehicles in a scenic area according to the embodiments of the present disclosure.

[0106] Those skilled in the art will appreciate that the contents disclosed in this disclosure may be subject to various variations and improvements. For example, the various devices or components described above may be implemented by hardware, or by software, firmware, or a combination of some or all of the three.

[0107] In addition, although the present disclosure makes various references to certain units in the system according to embodiments of the present disclosure, any number of different units may be used and run on the client and / or server. The units are illustrative only, and different aspects of the systems and methods may use different units.

[0108] Those skilled in the art will appreciate that all or part of the steps in the above method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk. Optionally, all or part of the steps in the above embodiment can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiment can be implemented in the form of hardware or in the form of a software functional module. The present disclosure is not limited to any particular form of combination of hardware and software.

[0109] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present disclosure belongs. It should also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology and should not be interpreted in an idealized or highly formal sense, unless explicitly defined as such herein.

[0110] The above is an explanation of the present disclosure and should not be considered as a limitation thereof. Although several exemplary embodiments of the present disclosure have been described, it will be readily understood by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present disclosure. Therefore, all such modifications are intended to be included within the scope of the present disclosure as defined in the claims. It should be understood that the above is an explanation of the present disclosure and should not be considered as being limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present disclosure is defined by the claims and their equivalents.

Claims

1. A smart diversion system for people and vehicles in scenic spots, It is characterized in that include: A crowd monitoring video acquisition module is used to obtain crowd monitoring videos collected by cameras deployed at intersections in scenic spots; A crowd flow monitoring key frame sampling module is used to perform key frame discrete sampling on the crowd flow monitoring video to obtain a time series of crowd flow monitoring key frames; A crowd distribution semantic feature extraction module, used for extracting features from each crowd flow monitoring key frame in the time series of the crowd flow monitoring key frames through a crowd flow distribution feature extractor based on a deep neural network model to obtain a time series of crowd flow distribution semantic feature graphs; A crowd flow distribution semantic spatial feature attention module is used to obtain a time series of spatially significant crowd flow distribution semantic feature maps by passing the time series of the crowd flow distribution semantic feature maps through a spatial information multi-scale intensive retention attention module; A pedestrian flow time series reasoning module, used for passing the time series of the spatially significant pedestrian flow distribution semantic feature graph through the pedestrian flow time series distribution reasoning module to obtain a pedestrian flow time series reasoning semantic feature vector as a pedestrian flow time series reasoning semantic feature; as well as The green light time control module is used to determine whether to extend the green light time based on the semantic features of the pedestrian flow temporal reasoning.

2. The intelligent traffic diversion system for people and vehicles in scenic spots according to claim 1, It is characterized in that The deep neural network model is a convolutional neural network model.

3. The intelligent traffic diversion system for people and vehicles in scenic spots according to claim 2, It is characterized in that The crowd flow distribution semantic space feature attention module includes: A crowd distribution semantic feature map multi-mode pooling unit, used for performing multi-mode pooling processing on each crowd distribution semantic feature map in the time series of the crowd distribution semantic feature map along the channel dimension to obtain a spliced ​​crowd distribution semantic feature matrix; A crowd flow distribution implicit correlation feature capturing unit, used for performing convolution processing on the spliced ​​crowd flow distribution semantic feature matrix through a convolution layer to obtain a crowd flow distribution implicit correlation feature matrix; A crowd flow distribution spatial attention weight calculation activation unit, used for inputting the crowd flow distribution implicit correlation feature matrix into a Sigmoid activation unit to obtain a crowd flow distribution spatial attention weight matrix; and The spatial saliency enhancement unit is used to multiply each feature matrix along the channel dimension in the crowd distribution semantic feature map by the crowd distribution spatial attention weight matrix according to the position point to obtain the spatially salient crowd distribution semantic feature map.

4. The intelligent traffic diversion system for people and vehicles in scenic spots according to claim 3, It is characterized in that The crowd flow distribution semantic feature map multi-mode pooling unit includes: A global mean pooling subunit, used for performing global mean pooling processing on the crowd flow distribution semantic feature map along the channel dimension to obtain a crowd flow distribution semantic global mean pooling feature matrix; A maximum pooling subunit, used for performing maximum pooling processing on the crowd flow distribution semantic feature map along the channel dimension to obtain a crowd flow distribution semantic maximum pooling feature matrix; A random pooling subunit, configured to perform random pooling processing on the crowd flow distribution semantic feature map along the channel dimension to obtain a crowd flow distribution semantic random pooling feature matrix; and The multi-pooling semantic splicing unit is used to splice the crowd distribution semantic mean pooling feature matrix, the crowd distribution semantic maximum pooling feature matrix and the crowd distribution semantic random pooling feature matrix to obtain the spliced ​​crowd distribution semantic feature matrix.

5. The intelligent traffic diversion system for people and vehicles in scenic spots according to claim 4, It is characterized in that The pedestrian flow temporal reasoning module includes: A vectorization expansion unit, used for vectorizing and expanding a plurality of spatially significant crowd flow distribution semantic feature graphs in the time series of the spatially significant crowd flow distribution semantic feature graphs to obtain a sequence of spatially significant crowd flow distribution semantic feature vectors; A vector screening unit, used to determine the spatially significant crowd flow distribution semantic feature vector at the last position in the sequence of the spatially significant crowd flow distribution semantic feature vectors; A crowd flow distribution average information difference calculation unit is used to calculate the position difference between the mean feature vector of multiple spatially significant crowd flow distribution semantic feature vectors in the sequence of the spatially significant crowd flow distribution semantic feature vectors and the spatially significant crowd flow distribution semantic feature vector at the last position to obtain a crowd flow distribution average information difference feature vector; A random information difference calculation unit for crowd flow distribution is used to calculate the position difference between the spatially significant crowd flow distribution semantic feature vector at a random position in the sequence of the spatially significant crowd flow distribution semantic feature vector and the spatially significant crowd flow distribution semantic feature vector at the last position to obtain a random information difference feature vector for crowd flow distribution; A first crowd flow distribution inference unit, configured to multiply the crowd flow distribution average information difference feature vector by a first predetermined weight hyperparameter and a first Gaussian distribution random number function value to obtain a first crowd flow distribution inference feature vector; A second crowd flow distribution inference unit, configured to multiply the crowd flow distribution random information difference feature vector by a second predetermined weight hyperparameter and a second Gaussian distribution random number function value to obtain a second crowd flow distribution inference feature vector; and A fusion unit is used to calculate the positional addition between the first crowd flow distribution reasoning feature vector and the second crowd flow distribution reasoning feature vector to obtain the crowd flow temporal reasoning semantic feature vector.

6. The intelligent traffic diversion system for people and vehicles in scenic spots according to claim 5, It is characterized in that The first Gaussian distribution random number function value and the second Gaussian distribution random number function value are generated by a Gaussian distribution random number function with a mean of 0 and a variance of 1.

7. The intelligent traffic diversion system for people and vehicles in scenic spots according to claim 6, It is characterized in that The green light time control module is used to: The pedestrian flow temporal reasoning semantic feature vector is passed through a traffic light controller based on a classifier to obtain a control instruction, wherein the control instruction is used to indicate whether to extend the green light time.

8. The intelligent traffic diversion system for people and vehicles in scenic spots according to claim 7, It is characterized in that It also includes a training module for training a crowd distribution feature extractor based on a convolutional neural network model, the spatial information multi-scale intensive retention attention module, the crowd flow temporal distribution reasoning module and the classifier-based traffic light controller.

9. The intelligent traffic diversion system for people and vehicles in scenic spots according to claim 8, It is characterized in that The training module comprises: A training data acquisition unit, used to acquire training data, wherein the training data includes a training flow monitoring video collected by a camera deployed at a scenic spot intersection; A training crowd flow monitoring key frame sampling unit, used for performing key frame discrete sampling on the training crowd flow monitoring video to obtain a time series of training crowd flow monitoring key frames; A training crowd flow distribution semantic feature extraction unit, used for extracting features from each training crowd flow monitoring key frame in the time series of the training crowd flow monitoring key frames through the crowd flow distribution feature extractor based on the convolutional neural network model to obtain a time series of training crowd flow distribution semantic feature graphs; A training crowd flow distribution semantic space feature attention unit is used to pass the time series of the training crowd flow distribution semantic feature map through the spatial information multi-scale dense retention attention module to obtain a time series of training space-significant crowd flow distribution semantic feature map; A training pedestrian flow time series reasoning unit is used to pass the time series of the training spatially significant pedestrian flow distribution semantic feature graph through the pedestrian flow time series distribution reasoning module to obtain a training pedestrian flow time series reasoning semantic feature vector; A loss calculation unit, used for passing the training pedestrian flow temporal reasoning semantic feature vector through the classifier-based traffic light controller to obtain a classification loss function value; and A loss training unit is used to train the crowd distribution feature extractor based on the convolutional neural network model, the spatial information multi-scale intensive retention attention module, the crowd flow temporal distribution reasoning module and the classifier-based traffic light controller based on the classification loss function value, wherein in each round of iteration of the training, the training crowd flow temporal reasoning semantic feature vector is corrected.

10. A method for intelligently directing people and vehicles in scenic spots. It is characterized in that include: Obtain the crowd monitoring video collected by the cameras deployed at the intersections of the scenic area; Performing key frame discrete sampling on the crowd flow monitoring video to obtain a time series of crowd flow monitoring key frames; By using a crowd distribution feature extractor based on a deep neural network model, feature extraction is performed on each of the crowd flow monitoring key frames in the time series of the crowd flow monitoring key frames to obtain a time series of crowd flow distribution semantic feature graphs; The time series of the semantic feature map of crowd flow distribution is passed through a spatial information multi-scale dense retention attention module to obtain a time series of spatially significant semantic feature maps of crowd flow distribution; The time series of the spatially significant crowd flow distribution semantic feature graph is passed through a crowd flow temporal distribution reasoning module to obtain a crowd flow temporal reasoning semantic feature vector as a crowd flow temporal reasoning semantic feature; as well as Based on the pedestrian flow time series reasoning semantic features, determine whether to extend the green light time.

Images