Pedestrian flow intelligent analysis method based on subway station video monitoring

Abstract

An intelligent pedestrian flow analysis method based on subway station video monitoring comprises the following steps: setting a decomposition frequency to continuously decompose and screen station video data to obtain a screening frame element library, setting a linear activation kernel and carrying out convergence extraction to obtain a coverage feature value, and carrying out element replacement on the linear activation kernel according to the coverage feature value to obtain a pedestrian flow analysis result; obtaining a replacement capture core; carrying out mark capture and nuclide matching on the screening frame element library by replacing the capture kernel to obtain a net value matching kernel, and constructing an element net value library according to the net value matching kernel; carrying out identity analogy on the element net value library to obtain a similar identity net value library, carrying out pattern positioning on the net value matching core and carrying out similar identification on the net value matching core and the similar identity net value library to obtain a target class identification node, and carrying out position tracking and heterogeneous analysis on the target identification node to obtain a traffic adjustment strategy; the passenger flow is automatically detected and tracked, and the passenger flow management efficiency of the subway station is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent video surveillance, and particularly to an intelligent analysis method for the number of passengers based on video surveillance in subway stations. Background Art

[0002] As a large-capacity and high-efficiency urban transportation means, the subway plays an important role in many big cities. However, the high passenger flow in subway stations also brings many management problems.

[0003] In order to better understand the changing rules of passenger flow, reasonably arrange vehicle schedules, shorten or extend the departure intervals, it is necessary to monitor and analyze the number of passengers in real time, so as to improve the operation efficiency and passenger satisfaction; during peak hours, there may be congestion in subway stations, and intelligent analysis can help the operation side take timely measures to guide the flow of passengers and prevent the occurrence of stampede accidents; understanding the passenger flow distribution in subway stations helps the subway operation side reasonably allocate human resources and equipment resources.

[0004] Therefore, in view of the above defects, the designers of the present invention, through painstaking research and design, integrating the experience and achievements of being engaged in related industries for many years, have studied and designed an intelligent analysis method for the number of passengers based on video surveillance in subway stations to overcome the above defects. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent analysis method for the number of passengers based on video surveillance in subway stations, which can effectively overcome the defects of the prior art, improve the operation efficiency of subway stations, ensure passenger safety, and reduce operation costs.

[0006] To achieve the above purpose, the present invention discloses an intelligent analysis method for the number of passengers based on video surveillance in subway stations, which is characterized by including the following steps: Step S1: Construct a station layout diagram and collect station video data; Step S2: Set the decomposition frequency to continuously decompose the station video data to obtain continuous video frames, construct a continuous frame element library, conduct an initial screening on the continuous frame element library to obtain a screened frame element library, set a linear activation kernel and upload it to the marked frame elements to obtain a core coverage area, conduct a convergence extraction on the core coverage area to obtain a coverage feature value, and replace the elements of the linear activation kernel according to the coverage feature value to obtain a replacement capture kernel; Step S3: Conduct a mark capture on the screened frame element library through the replacement capture kernel to obtain feature replacement elements, conduct a point position sorting on the feature replacement elements and conduct a nuclide matching with the replacement capture kernel to obtain a net value matching kernel, and construct an element net value library according to the net value matching kernel; Step S4: Locate the pattern of the net worth matching core to obtain the target mapping node, conduct identity analogy on the element net worth library to obtain the same-class identity net worth library, use the same-class identity net worth library to perform the same-class identification on the target mapping node to obtain the target class identification node, and conduct position tracking and heterogeneous analysis on the target identification node to obtain the traffic regulation strategy.

[0007] Among them, the process of collecting station video data includes: Construct a station layout map based on the subway station; Conduct monitoring and defense on the subway station to obtain the video acquisition end; Conduct temporal acquisition through the video acquisition end to obtain the station video data, and perform time marking on the collected station video data to obtain the synchronous time point.

[0008] Among them, the process of obtaining the screened frame element library includes: Set the decomposition frequency, continuously decompose the station video data according to the decomposition frequency to obtain continuous video frames; Based on the synchronous time point, perform time association on the continuous video frames according to the decomposition frequency to obtain frame element nodes; Based on the frame element nodes, construct a continuous frame element library from the continuous video frames, upload the obtained continuous video frames to the continuous frame element library, and conduct initial screening on the continuous frame element library to obtain the screened frame element library, where the screened frame element library includes marked frame elements.

[0009] Among them, the process of obtaining the replacement capture core includes: Perform pixel recognition on the obtained screened element library to obtain frame image pixel points; Set a linear activation core according to the marked frame elements, select the center of the marked frame elements to obtain the center pixel points, upload the linear activation core to the marked frame elements based on the center pixel points, and mark the corresponding area of the linear activation core in the marked frame elements as the core coverage area; Conduct confluence extraction on the obtained core coverage area to obtain the coverage feature value, and perform element replacement on the linear activation core according to the coverage feature value to obtain the replacement capture core.

[0010] Among them, the process of conducting confluence extraction on the core coverage area includes: Based on the core coverage area, perform in-region expansion on the frame image pixel points to obtain the in-point coverage area, and perform row and column extraction on the in-point coverage area to obtain the partial derivative coverage area; Perform feature calculation on the obtained partial derivative coverage area to obtain the coverage feature value.

[0011] Among them, the process of using the replacement capture core to perform marker capture on the screened frame element library to obtain the feature replacement element includes: Set the capture origin according to the marked frame element, upload the replacement capture core to the capture origin of the marked frame element based on the filtered frame element library, and mark the corresponding area of the replacement capture core within the marked frame element as the replacement comparison area; Extract the section of the replacement comparison area through the replacement capture core to obtain the extraction comparison area; Set the migration step size according to the replacement capture core, and traverse and mobilize the replacement capture core based on the marked frame element according to the migration step size to reach the next replacement comparison area, repeat the process of obtaining the extraction comparison area until all areas of the marked frame element are covered, and traverse and combine the extraction comparison areas according to the order of the traversal and mobilization to obtain the feature replacement element.

[0012] Among them, the process of constructing the element net value library according to the net value matching core includes: Sort the point positions of the feature replacement elements to obtain the net value sequence; Perform nuclide matching on the net value sequence according to the obtained replacement capture core to obtain the net value matching core; Construct the element net value library according to the net value matching core, and upload the obtained net value matching core to the element net value library.

[0013] Among them, the process of obtaining the same-class identity net value library includes: Obtain the video acquisition end corresponding to the net value matching core, and perform temporal positioning on the net value matching core according to the obtained video acquisition end to obtain the net value position node; Map the net value matching core to the station layout diagram according to the net value position node to obtain the target mapping node; Perform identity analogy on the element net value library based on the station layout diagram to obtain the same-class identity net value library.

[0014] Among them, the process of performing identity analogy on the element net value library includes: Select an arbitrary net value matching core in the element net value library as the comparison matching core, compare the obtained comparison matching core with the net value matching core to obtain the determination result core; Perform same-class supplementation on the comparison matching core based on the element net value library according to the determination result core to obtain the same-class identity net value library; Perform multi-source screening on the element net value library based on the same-class identity net value library to obtain the filtered net value library, select an arbitrary net value matching core in the filtered net value library as the comparison matching core, and repeat the process of obtaining the same-class identity net value library until all net value matching cores in the element net value library have completed type assignment.

[0015] Among them, the process of obtaining the passage adjustment strategy includes: Perform same-class identification on the target mapping node based on the same-class identity net value library according to the station layout diagram to obtain the target class identification node; Set the observation period according to the synchronization time point, track the position of the target class identification node based on the observation period to obtain the target period trace, and perform heterogeneous analysis on the target period trace to obtain the traffic regulation strategy.

[0016] As can be seen from the above, an intelligent analysis method for the passenger flow based on the video surveillance of subway stations of the present invention has the following effects: 1. Conduct video surveillance and collection on subway stations, decompose the video data into frame images, identify the image characters, and extract the feature information of the frame images as the identity discrimination points for trace tracking, which greatly improves the data processing efficiency and the tracking speed of the target person.

[0017] 2. Construct a station layout map with the same scale as the subway station to display the real-time positions of pedestrians and obtain the periodic traces of the target persons through the identity discrimination points, which is beneficial to counting the passenger flow of popular stations and quickly generating corresponding traffic regulation strategies when the passenger flow of popular stations is too large, effectively relieving the passenger flow pressure.

[0018] The detailed content of the present invention can be obtained through the following description and the accompanying drawings. Brief Description of the Drawings

[0019] Figure 1 Shows a schematic diagram of an intelligent analysis method for the passenger flow based on the video surveillance of subway stations of the present invention. Detailed Embodiments

[0020] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0021] As Figure 1 shown, shows an intelligent analysis method for the passenger flow based on the video surveillance of subway stations of the present invention, and the method includes the following steps: Step S1: Construct a station layout map and collect station video data; Step S2: Set the decomposition frequency to continuously decompose the station video data to obtain continuous video frames, construct a continuous frame element library, perform an initial screening on the continuous frame element library to obtain a screened frame element library, set a linear activation kernel and upload it to the marked frame elements to obtain a core coverage area, perform a convergence extraction on the core coverage area to obtain a coverage feature value, and replace the elements of the linear activation kernel according to the coverage feature value to obtain a replacement capture kernel; Step S3: Perform flag capture by replacing and capturing the check and screening frame element library to obtain feature replacement elements. Sort the positions of the feature replacement elements and perform nuclide matching with the replacement capture core to obtain a net value matching core. Construct an element net value library based on the net value matching core; Step S4: Locate the pattern of the net value matching core to obtain a target mapping node. Conduct an identity analogy on the element net value library to obtain a same-class identity net value library. Use the same-class identity net value library to perform same-class identification on the target mapping node to obtain a target class identification node. Conduct position tracking and heterogeneous analysis on the target identification node to obtain a traffic regulation strategy.

[0022] It should be further noted that in the specific implementation process, the process of collecting station video data in step S1 includes: Construct a station layout diagram based on the subway station. The station layout diagram is a three-dimensional floor plan generated according to the actual structure of the subway station. The positions of the equipment and facilities in the station layout diagram are exactly the same as those of the subway station; Set up monitoring defenses for the subway station to obtain a video acquisition terminal. The station defense point represents the acquisition of the overall structure of the subway station. Set up station defense points at the positions where monitoring is required to set up monitoring equipment for information collection, which is the video acquisition terminal; Map the obtained video acquisition terminal to the station layout diagram to obtain a station defense point. Among them, the station defense point is the corresponding position of the video acquisition terminal represented in the station layout diagram; Construct a transmission link between the video acquisition terminal and the station defense point. The transmission link is used to transmit the information collected by the video acquisition terminal to the station defense point; Perform temporal acquisition through the video acquisition terminal to obtain station video data, and mark the collected station video data with time to obtain a synchronous time point; Furthermore, the time marking means marking the starting point of the collected station video data, marking the ending time point of the collection, and at the same time, marking the video duration of the station video data, that is, the corresponding time point can be obtained at any moment of the station video data.

[0023] Among them, the decomposition frequency in step S2 means setting the frequency of decomposing the station video data. The decomposition means decomposing the station video data into several consecutive image forms; Continuously decompose the station video data according to the obtained decomposition frequency to obtain continuous video frames; Further, the continuous decomposition representation segments each frame of the station video data according to the obtained decomposition frequency, that is, uploading the obtained station video data to the video processing library software, reading the station video data file through the video processing library software, finding the corresponding segmented frames according to the required segmentation frequency distribution, and denoting the corresponding segmented frames as continuous video frames; in particular, since the video data is composed of continuous frames, the continuous video frames obtained by segmenting and extracting frames according to the decomposition frequency are in the form of images; Time-correlate the continuous video frames according to the obtained decomposition frequency based on the synchronous time points to obtain frame meta-nodes. Among them, the time-correlation means that since the station video data has corresponding synchronous time points, each frame in the station video data is composed in the time sequence of the synchronous time points, so that the frame meta-node corresponding to each continuous video frame can be obtained; Construct a continuous frame element library based on the frame meta-nodes according to the obtained continuous video frames, and upload the obtained continuous video frames to the continuous frame element library in the order of the frame meta-nodes, that is, the arrangement order of the continuous video frames in the continuous frame element library is arranged in the chronological order; Perform an initial screening on the obtained continuous frame element library to obtain a screened frame element library, and the screened frame element library includes several marked frame elements; Further, the process of the initial screening includes: Perform human portrait marking on the continuous frame element library to obtain a marked frame element library. The human portrait marking means that in the continuous frame element library, pedestrians appearing in the continuous video frames are marked through a pedestrian detection algorithm, that is, the continuous video frames where pedestrians appear are denoted as marked frame elements, and the continuous video frames where no pedestrians are detected are denoted as invalid frame elements until all the continuous video frames in the continuous frame element library are marked to obtain a marked frame element library. Among them, the pedestrian detection algorithm includes but is not limited to the HOG feature + SVM algorithm, DPM algorithm, YOLO algorithm, SSD algorithm, Faster R-CNN algorithm; Screen the marked frame element library according to the obtained invalid frame elements, that is, remove the invalid frame elements in the marked frame element library and retain the marked frame elements to obtain a screened frame element library; Perform pixel recognition on the obtained screened element library to obtain frame image pixel points; The pixel recognition means marking the pixel point positions in the marked frame elements in the screened element library to obtain frame image pixel points, denoting the obtained frame image pixel points as (x, y), and denoting the pixel value at the frame image pixel points as the pixel normalization value; Set a linear activation kernel according to the obtained marked frame elements. The linear activation kernel is in the form of a matrix, and the number of rows and columns of the matrix is equal, denoted as a matrix of i rows and j columns, that is, i = j, and the matrix is composed of elements inside; In particular, the elements within the set linear activation kernel can correspond to the frame image pixels within the marked frame element at corresponding positions, that is, the linear activation kernel can adapt to the arrangement of the frame image pixels within the marked frame element; Centrally select the obtained marked frame element to obtain a central pixel point, where the central pixel point represents the pixel point at the center of the marked frame element; Based on the central pixel point, upload the obtained linear activation kernel to the marked frame element, and denote the corresponding area of the linear activation kernel within the marked frame element as the core coverage area, that is, make the element in the i-th row and j-th column of the linear activation kernel coincide with the central pixel point, which means the upload is successful, and the frame image pixels within the core coverage area correspond one-to-one with the elements within the linear activation kernel; then the manifestation form of the core coverage area is the same as that of the linear activation kernel, that is, the number of frame image pixels within the core coverage area is the same as the arrangement of the elements within the linear activation kernel, which is an area composed of frame image pixels in i rows and j columns; Perform confluence extraction on the obtained core coverage area to obtain a coverage eigenvalue, and replace the elements of the linear activation kernel according to the coverage eigenvalue to obtain a replacement capture kernel; It should be further noted that in the specific implementation process, the process of the confluence extraction includes: Mark the frame image pixels within the core coverage area, denoted as , where (i, j) represents the frame image pixel at the i-th row and j-th column of the core coverage area; Perform in-region expansion on the frame image pixel at the first row and first column within the core coverage area to obtain an in-point coverage area; The in-region expansion means taking the frame image pixel at the first row and first column of the core coverage area as the center, covering the surrounding elements to form an i-row and j-column matrix area centered on the frame image pixel at the first row and first column, denoted as the in-point coverage area, that is, the in-point coverage area is also composed of frame image pixels in i rows and j columns, with the frame image pixel at the first row and first column within the core coverage area as the center and this center being the element at the i-th row and j-th column of the in-point coverage area; Perform row-column extraction on the obtained in-point coverage area to obtain a partial derivative coverage area; The process of the row-column extraction includes: Calculate the first-order partial derivative of the frame image pixels within the in-point coverage area in the horizontal direction, that is, calculate the first-order partial derivative in the x direction, and denote the obtained matrix as the partial row coverage area; Then calculate the first-order partial derivative of the frame image pixels within the in-point coverage area in the vertical direction, that is, calculate the first-order partial derivative in the y direction, and denote the obtained matrix as the partial column coverage area; Calculate the second-order partial derivative of the partial column coverage area according to the obtained partial row coverage area, that is, calculate the y-direction partial derivative of the partial row coverage area and the x-direction partial derivative of the partial column coverage area, which is the second-order partial derivative, and denote it as the partial derivative coverage area; Feature extraction is performed on the obtained partial derivative coverage area to obtain a coverage feature value. Here, feature extraction means performing a determinant operation on the obtained partial derivative coverage area, that is, calculating the determinant value of the partial derivative coverage area, and marking the obtained determinant value as the coverage feature value; And so on, perform in-region expansion on the frame image pixel points in the 2nd column of the 1st row in the core coverage area, repeat the process of obtaining the coverage feature value, continue to perform in-region expansion on the frame image pixel points in the 3rd column of the 1st row in the core coverage area, perform in-region expansion on the frame image pixel points in the 4th column of the 1st row in the core coverage area... until all frame image pixel points in the core coverage area have obtained corresponding coverage feature values; Element replacement is performed on the obtained linear activation kernel according to the obtained coverage feature value to obtain a replacement capture kernel; Further, the process of the element replacement includes: Since the coverage feature value is obtained corresponding to the frame image pixel points in the 1st column of the 1st row, the 2nd column of the 1st row, the 3rd column of the 1st row, the 4th column of the 1st row... in the corresponding core coverage area, the element in the 1st column of the 1st row in the linear activation kernel is replaced with the coverage feature value corresponding to the frame image pixel point in the 1st column of the 1st row in the core coverage area, the element in the 2nd column of the 1st row in the linear activation kernel is replaced with the coverage feature value corresponding to the frame image pixel point in the 2nd column of the 1st row in the core coverage area, until all elements in the linear activation kernel are replaced to obtain a replacement capture kernel.

[0024] The obtained replacement capture kernel can be uploaded to the screening frame element library, and the screening frame element library is marked and captured through the replacement capture kernel to obtain feature replacement elements; It should be further noted that in the specific implementation process, among them, the process of the marker capture in step S3 includes: Set a capture origin according to the obtained marked frame element, and upload the obtained replacement capture kernel to the capture origin of the marked frame element based on the screening frame element library. Here, the capture origin refers to the point set in the marked frame element for determining the initial movement position of the replacement capture kernel, that is, upload the replacement capture kernel to the first marked frame element in the screening frame element library, and mark the corresponding area of the replacement capture kernel in the marked frame element as the replacement comparison area; Perform section extraction on the replacement comparison area through the replacement capture kernel to obtain an extraction comparison area; Further, the section extraction means convolving the elements in the replacement capture kernel with the pixel normalization values of the corresponding frame image pixel points in the replacement comparison area, and convolving with the elements at the corresponding positions to obtain the extraction comparison area, and then replacing the pixel normalization values of the frame image pixel points in the extraction comparison area with the values after convolution; Set the migration step according to the obtained replacement capture kernel. The migration step represents the moving distance for moving the replacement capture kernel within the marked frame element, that is, the migration step. In order to improve the accuracy of feature extraction of the marked frame element, the set migration step needs to be small enough, that is, the migration step is equal to half of the number of rows of the replacement capture kernel, that is, each time it moves the distance of a frame image pixel point; Based on the marked frame element, traverse and mobilize the obtained replacement capture kernel according to the obtained migration step, reach the next replacement comparison area, and perform section extraction on the replacement comparison area through the replacement capture kernel until all areas of the marked frame element are covered. Traverse and combine the obtained extraction comparison areas according to the order of traversal and mobilization to obtain feature replacement elements. Then, the obtained feature replacement elements have the same form as the marked frame elements, the positions of the frame image pixel points remain unchanged, only the pixel normalization values at the frame image pixel points change, which are the values after convolution with the replacement capture kernel, and record the convolved pixel normalization values as replacement normalization values; Further, the traversal and mobilization means moving the replacement capture kernel at the capture origin position by the distance of the migration step to reach the next replacement comparison area, and performing section extraction with the replacement comparison area to obtain an extraction comparison area, and the order of traversal and mobilization is from left to right. After reaching the boundary of the marked frame element, move down to the position of the height of one replacement capture kernel and continue to move to the right; Specifically, if the replacement capture kernel exceeds the boundary of the marked frame element after moving the distance of the migration step, it only moves to the position where the replacement capture kernel coincides with the boundary of the marked frame element; Perform point position sorting on the obtained feature replacement elements to obtain a net value sequence; The point position sorting means sorting the values of the frame image pixel points in the feature replacement elements in descending order to obtain a net value sequence, and the net value sequence arranges the frame image pixel points; Perform nuclide matching on the net value sequence according to the obtained replacement capture kernel to obtain a net value matching kernel; The nuclide matching means constructing a matrix of i rows and j columns according to the arrangement order of the replacement capture kernel, that is, a matrix of i rows and j columns, denoted as the original sequence kernel, and the constructed matrix also has the same number of rows and columns. Select the top i×j frame image pixel points in the net value sequence, and upload the selected frame image pixel points to the constructed original sequence kernel in sequence, and retain the corresponding values of the frame image pixel points at the corresponding element positions of the original sequence kernel to obtain a net value matching kernel. Then, the obtained net value matching kernel is a matrix form composed of the replacement normalization values of the top i×j frame image pixel points in the corresponding net value sequence; Associate the obtained net value matching kernel with the corresponding marked frame element; Construct an element net value library according to the obtained net value matching kernel, and upload the obtained net value matching kernel to the element net value library.

[0025] Obtain the video acquisition terminal corresponding to the net value matching kernel, perform temporal positioning on the net value matching kernel according to the obtained video acquisition terminal, and obtain the net value position node; Among them, the temporal positioning described in step S4 means that according to the station video data corresponding to the net value matching kernel, the video acquisition terminal corresponding to the station video data can be obtained. The position information corresponding to the net value matching kernel is determined through the monitoring and defense position of the video acquisition terminal in the subway station, denoted as the net value position node, and the obtained net value position node is associated with the corresponding net value matching kernel; Map the obtained net value matching kernel to the station layout diagram according to the obtained net value position node to obtain the target mapping node, and upload the obtained net value matching kernel to the corresponding target mapping node; The target mapping node means uploading the net value position node to the station layout diagram through the transmission link and mapping it to the same position to obtain the target mapping node. There is a corresponding associated net value matching kernel at the target mapping node, that is, the net value matching kernel represents the identity characteristic information corresponding to the marked frame element, and all the net value position nodes corresponding to the obtained net value matching kernels are mapped to the station layout diagram; Perform identity analogy on the obtained element net value library based on the station layout diagram to obtain the same-class identity net value library; It should be further noted that in the specific implementation process, the process of the identity analogy includes: Select a net value matching kernel in the element net value library as the reference matching kernel, select a net value matching kernel from the remaining net value matching kernels, compare the obtained reference matching kernel with the net value matching kernel to obtain the judgment result kernel; Furthermore, the process of the difference comparison includes: Perform difference operation on the corresponding position elements in the reference matching kernel and the net value matching kernel, that is, subtract the corresponding position elements in the reference matching kernel and the net value matching kernel to obtain the analogy kernel; Perform element judgment on the obtained analogy kernel to obtain the judgment result kernel, and the judgment result kernel includes the matching type kernel and the non-matching type kernel; The element judgment means judging each element in the analogy kernel one by one. When all elements in the analogy kernel are zero, the analogy kernel is denoted as the matching type kernel. When there are elements in the analogy kernel that are not all zero, the analogy kernel is denoted as the non-matching type kernel; Select another net value matching kernel from the remaining net value matching kernels and perform difference comparison with the reference matching kernel until all the remaining net value matching kernels have completed difference comparison with the reference matching kernel; Based on the element net value library, perform same-class supplement on the reference matching kernel according to the judgment result kernel to obtain the same-class identity net value library; The said homogeneous supplement indicates obtaining the judgment result core after differential comparison with the control matching library. When the judgment result core is a matching type core, a homogeneous identity net value library is constructed according to the net value matching core corresponding to the matching type core and the control matching core, and the net value matching core corresponding to the matching type core and the control matching core are uploaded to the homogeneous identity net value library. That is, the person identities within the marker frame elements corresponding to the net value matching core of the matching type core and the control matching core are the same pedestrian. That is, the pedestrian identities in the homogeneous identity net value library are exactly the same as the pedestrian information of the control matching core, except that at different times, the captured positions and different behaviors are involved. Based on the homogeneous identity net value library, multi-source screening is performed on the element net value library to obtain a filtered net value library. The said multi-source screening means removing the net value matching cores included in the homogeneous identity net value library from the element net value library. Select any net value matching core in the obtained filtered net value library as the control matching core, and repeat the process of differential comparison for the control matching core until all the net value matching cores in the element net value library have been differentially compared, completing the type assignment. The said type assignment means that all the net value matching cores in the element net value library can be matched to the corresponding matching type core, that is, they have the same homogeneous identity net value library. That is, one homogeneous identity net value library represents the image information of a pedestrian in a subway station within the duration of the video data of a station. Based on the station layout diagram, homogeneous identification is performed on the target mapping nodes according to the obtained homogeneous identity net value library to obtain target class identification nodes. The said homogeneous identification means, in the station layout diagram, according to the homogeneous identity net value library after type assignment, marking the target mapping nodes associated with the net value matching cores in the same homogeneous identity net value library as the same identification points, denoted as target class identification nodes. For example, marking them the same color, then in the station layout diagram, it is possible to distinguish whether pedestrians are the same person by observing the color of the target mapping nodes. Set an observation period according to the obtained synchronization time point. The said observation period represents a relatively long time period. According to the operation rules and monitoring requirements of the subway station, combined with the synchronization time point, set a relatively long time period as the observation period, and this time period should be able to cover multiple passenger flow peak and trough periods. Based on the observation period, position tracking is performed on the target class identification nodes to obtain target period trajectories. It should be further noted that in the specific implementation process, the process of the said position tracking includes: Match the time sequence of the frame element corresponding to the net value of the target class identification node with the frame element node corresponding to the nuclear marking frame, and perform virtual connection on the target class identification nodes in the station layout map to obtain the action trajectory of the target pedestrian corresponding to the same-class identity net value library in the subway station, which is the target periodic trajectory, representing the action trajectory in the subway station within a period of time. By knowing the riding whereabouts of the target pedestrian through the station layout map, where the target pedestrian refers to the pedestrian corresponding to a same-class identity net value library under study; Perform heterogeneous analysis on the obtained target periodic trajectory to obtain a traffic regulation strategy; The heterogeneous analysis refers to analyzing the path coincidence of the target periodic trajectories corresponding to different same-class identity net value libraries; Calculate the coincidence degree of the target periodic trajectories corresponding to different same-class identity net value libraries based on the observation period to obtain the trajectory coincidence degree. The trajectory coincidence degree represents counting the repeated target class identification nodes in all target periodic trajectories within an observation period to obtain the number of coincidence nodes. Among them, the number of coincidence nodes represents the number of times the same target class identification node appears in different target periodic trajectories; Set a node number threshold according to the obtained number of coincidence nodes, and generate a traffic regulation strategy through the node number threshold; Then relieve the traffic flow pressure at the net value position node according to the obtained traffic regulation strategy. The node number threshold represents the threshold number set according to the real-time operation situation of the subway station and is used to generate the traffic regulation strategy. The traffic regulation strategy means that when the number of coincidence nodes of a target identification node is greater than or equal to the node number threshold, a diversion strategy is generated according to the number of coincidence nodes at other target identification nodes, and the crowd at this target identification node is dispersed to other target identification nodes to obtain the traffic regulation strategy.

[0026] Thus, the advantages of the present invention are as follows: By decomposing the collected video data into an image form, marking the pixel points in the image, and setting an extraction kernel, the central pixel point is used to generate a second-order partial derivative matrix to update the elements of the extraction kernel, and an alternative capture kernel for feature extraction is obtained, which is convenient for extracting various details in the image, enhancing the image content, and reducing the situations of misjudgment and missed judgment; Then, sliding capture is performed on the pixels in the updated capture check image to obtain the feature pixel values after feature extraction, and the feature pixel values are sorted to obtain a net value sequence. A net value matching kernel corresponding to each image is constructed based on the net value sequence. By setting key features for purposeful feature extraction, it helps to improve the reliability and accuracy of image feature extraction, enabling the computer to more accurately understand the information in the image, facilitating the management staff to set specific target features to search for target passengers in the station and increasing the retrieval speed. For example, if subway management staff need to find passengers wearing yellow hats and red backpacks, extraction kernels are set according to these two features of the yellow hat and the red backpack, and element updates are performed to obtain an alternated capture kernel to extract the relevant net value matching kernel in the image, accurately grasping the position information and behavior trajectory of the target passengers, helping the staff to make quick decisions and take effective emergency measures to ensure the safety of passengers. Based on the obtained net value matching kernels, a feature image library generated by each passenger in the subway station is constructed to obtain a filtered net value library, which contains all the image information of a passenger in the subway station within the duration of the video data of a station. The position nodes of each net value matching kernel in the filtered net value library are marked on the station layout diagram, and all the position nodes in the filtered net value library are connected according to the chronological order to obtain the action trajectory of the passenger within an observation period. This can more accurately discover the passenger flow patterns in the subway station, such as peak hours, popular areas, etc., which helps the subway operation department to reasonably arrange staff, adjust operation strategies, and improve operation efficiency. At the same time, according to the action trajectories of passengers in the subway station, in case of emergencies, clear images can provide accurate passenger flow information in a timely manner, helping the staff to make quick decisions and take effective emergency measures to ensure the safety of passengers. Finally, the repeated position nodes in the action trajectories of all passengers within an observation period are counted to obtain the number of overlapping nodes. The passenger flow volume of the position node is determined by the size of the number of overlapping nodes, and then a traffic regulation strategy is generated, providing a reference for the planning and optimization of the urban rail transit network, reasonably adjusting the line direction, station settings, etc., improving the coverage and service level of the rail transit network, and promoting the sustainable development of urban transportation.

[0027] Obviously, the above description and record are only examples and not intended to limit the disclosure, application, or use of the present invention. Although it has been described in the embodiments and illustrated in the drawings, the present invention does not limit to the specific examples described in the embodiments and illustrated in the drawings as the currently considered best mode for implementing the teachings of the present invention. The scope of the present invention will include any embodiments falling within the foregoing description and the appended claims.

Claims

1. An intelligent analysis method for passenger flow based on video surveillance of subway stations, characterized in that The following steps are involved: Step S1: construct a station layout map and collect station video data; Step S2: setting the decomposition frequency to continuously decompose the station video data, obtaining continuous video frames, constructing a continuous frame element library, initially screening the continuous frame element library, obtaining a screening frame element library, setting a linear activation kernel and uploading it to the marked frame element, obtaining a core coverage area, performing convergence extraction on the core coverage area, obtaining a coverage feature value, performing element replacement on the linear activation kernel according to the coverage feature value, and obtaining a replacement capture kernel; Step S3: perform mark capture on the screening frame element library through the alternating capture core to obtain characteristic replacement elements, sort the characteristic replacement elements by points and perform nuclide matching with the alternating capture core to obtain a net value matching core, and construct an element net value library according to the net value matching core; Step S4: perform pattern positioning on the net value matching core to obtain the target mapping node, perform identity analogy on the element net value library to obtain a similar identity net value library, perform similar identification on the target mapping node through the similar identity net value library to obtain the target class identification node, perform position tracking and heterogeneous analysis on the target identification node to obtain the traffic regulation strategy.

2. According to claim 1, a method for intelligent analysis of passenger flow based on subway station video surveillance is characterized in that: The process of collecting station video data includes: Construct station layout diagram based on subway stations; Monitor and deploy subway stations to obtain video acquisition terminals; Temporal acquisition is performed through the video acquisition terminal to obtain station video data, and the collected station video data is time-stamped to obtain the synchronization time point.

3. The method for intelligent analysis of passenger flow based on subway station video surveillance according to claim 1 is characterized in that: The process of obtaining the filter frame element library includes: Set the decomposition frequency, and continuously decompose the station video data according to the decomposition frequency to obtain continuous video frames; Based on the synchronization time point, the continuous video frames are temporally associated according to the decomposition frequency to obtain the frame element node; A continuous frame element library is constructed according to continuous video frames based on the frame element node, the obtained continuous video frames are uploaded to the continuous frame element library, and the continuous frame element library is initially screened to obtain a screening frame element library, wherein the screening frame element library includes marked frame elements.

4. The method for intelligent analysis of passenger flow based on subway station video surveillance according to claim 1 is characterized in that: The process of obtaining an alternate capture core includes: Perform pixel recognition on the obtained screening element library to obtain frame image pixel points; A linear activation kernel is set according to the marked frame element, the center of the marked frame element is selected to obtain the central pixel point, the linear activation kernel is uploaded to the marked frame element based on the central pixel point, and the corresponding area of ​​the linear activation kernel in the marked frame element is marked as the core coverage area; The obtained core coverage area is merged and extracted to obtain the coverage eigenvalue, and the linear activation kernel is replaced according to the coverage eigenvalue to obtain the replacement capture kernel.

5. The method for intelligent analysis of passenger flow based on video surveillance of subway stations according to claim 1 is characterized in that: The process of extracting the core coverage area includes: Based on the core coverage area, the pixel points of the frame image are expanded within the area to obtain the coverage area within the point, and the row and column extraction of the coverage area within the point is performed to obtain the partial derivative coverage area; The obtained partial derivative coverage area is characterized to obtain the coverage characteristic value.

6. The method for intelligent analysis of passenger flow based on subway station video surveillance according to claim 1 is characterized in that: The process of obtaining characteristic replacement elements by alternating capture and checking the screening frame element library for mark capture includes: Set the capture origin according to the marked frame element, upload the replacement capture core to the capture origin of the marked frame element based on the screening frame element library, and mark the corresponding area of ​​the replacement capture core in the marked frame element as the replacement control area; Performing segment extraction on the alternating control area by alternating capture check to obtain an extracted control area; The migration step is set according to the replacement capture core, and the replacement capture core is traversed and mobilized according to the migration step based on the marked frame element to reach the next replacement control area. The process of obtaining the extracted control area is repeated until all areas of the marked frame elements are covered. The extracted control areas are traversed and combined according to the order of traversal and mobilization to obtain the feature replacement elements.

7. The method for intelligent analysis of passenger flow based on video surveillance of subway stations according to claim 1 is characterized in that: The process of building an element net value library based on the net value matching core includes: Sort the characteristic replacement elements by points to obtain the net value sequence; Performing nuclide matching on the net value sequence according to the obtained replacement capture nucleus to obtain a net value matching nucleus; An element net value library is constructed according to the net value matching core, and the obtained net value matching core is uploaded to the element net value library.

8. The method for intelligent analysis of passenger flow based on video surveillance of subway stations according to claim 1 is characterized in that: The process of obtaining a homogeneous identity equity pool includes: Obtain the video acquisition end corresponding to the net value matching core, perform temporal positioning on the net value matching core according to the obtained video acquisition end, and obtain the net value position node; Map the net value matching core to the station layout diagram according to the net value position node to obtain the target mapping node; Based on the station layout diagram, the element net value library is compared with its identity to obtain a similar identity net value library.

9. The method for intelligent analysis of passenger flow based on video surveillance of subway stations according to claim 1 is characterized in that: The process of making identity analogies for the element net value library includes: A net value matching core is randomly selected in the element net value library and recorded as a reference matching core, and a difference comparison is performed between the obtained reference matching core and the net value matching core to obtain a determination result core; Based on the element net value library, check and compare the matching core according to the judgment results to supplement the same type and obtain the net value library of the same type of identity; Based on the net value library of the same identity, multi-source screening is performed on the element net value library to obtain the filtered net value library. Any net value matching core in the filtered net value library is selected as the control matching core, and the process of obtaining the net value library of the same identity is repeated until all net value matching cores in the element net value library have completed type assignment.

10. The method for intelligent analysis of passenger flow based on video surveillance of subway stations according to claim 1, characterized in that: The process of obtaining a traffic regulation strategy includes: Based on the station layout diagram, target mapping nodes are identified in the same category according to the same category identity net value library to obtain target category identification nodes; An observation period is set according to the synchronization time point, and the position of the target class identification node is tracked based on the observation period to obtain the target periodic trajectory. The target periodic trajectory is subjected to heterogeneous analysis to obtain the traffic regulation strategy.

Images