Subway passenger travel spatiotemporal trajectory recognition method and system based on pedestrian re-identification

Abstract

The present invention provides a method and system for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification. The method comprises the following steps: matching card swipe data from an automatic ticket vending and checking system, associating the card swipe data with surveillance video and image position to achieve binding between multiple categories of information; collecting a sample feature library of pedestrians from surveillance videos of each card number passing through entry and exit gates; searching for K travel paths along the entry-exit route; obtaining a valid set of alternative paths; calculating the earliest and latest times a passenger appears at each surveillance point; comparing the complete path set of each alternative path in the previous step to obtain a filtered set of comparison paths and a set of surveillance video nodes; calculating a minimum set selection scheme for surveillance nodes that meets preset requirements; and determining the path containing the surveillance node with the greatest feature similarity as the passenger's actual travel path. The method and system can efficiently achieve accurate passenger flow allocation results.

Description

Technical Field

[0001] The present invention belongs to the technical field of passenger flow monitoring, and more specifically, relates to a method and system for identifying the spatiotemporal trajectories of subway passengers based on pedestrian re-identification. More specifically, the present invention relates to a method, system, electronic device, and computer-readable storage medium for identifying the spatiotemporal trajectories of subway passengers based on pedestrian re-identification. Background Art

[0002] Large-scale networked operation of urban rail transit is the new normal for urban development. Thirteen cities, including Shanghai, Beijing, Chengdu, and Guangzhou, have operating mileage exceeding 300 km. The continued expansion of the network has led to a more complex network topology, more travel paths between any two nodes in the network, and more random passenger travel path selection behavior, posing a huge challenge to the accurate calculation of the urban rail transit network clearing and settlement model. Every year, major cities such as Beijing and Shanghai need to invest a large amount of manual surveys for parameter calibration and on-site verification of the clearing model to improve the model's calculation accuracy. The core challenge lies in how to accurately identify passengers' travel time and space paths in ultra-large-scale road networks to obtain accurate passenger flow distribution results. How to accurately identify passengers' travel time and space paths in ultra-large-scale road networks to obtain accurate passenger flow distribution results is a bottleneck problem that currently needs to be solved. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for identifying the spatiotemporal trajectories of subway passengers based on pedestrian re-identification, aiming to solve the technical problem of inaccurate identification of passengers' spatiotemporal travel paths in ultra-large-scale road networks in the prior art.

[0004] To achieve the above-mentioned object, the present invention adopts a technical solution of providing a method for identifying the spatiotemporal trajectories of subway passengers based on pedestrian re-identification, comprising the following steps:

[0005] Match card swiping data in the automatic ticket vending and checking system, associate card swiping data, surveillance video and screen position, and realize the binding between multiple categories of information;

[0006] Collect a sample feature library of pedestrians with each card number from surveillance videos at the entrance and exit gates;

[0007] Search for K travel paths along the inbound-outbound route; eliminate invalid paths to obtain a set of valid alternative paths;

[0008] Calculate the set of surveillance videos for any alternative route passing through each identification station; divide the alternative route into N segments, using transfer stations as the dividing points; and calculate the earliest and latest times a passenger appears at each monitoring point.

[0009] Compare the complete path set of each alternative path in the previous step, remove the overlapping identification stations, and obtain the filtered comparison path set and monitoring video node set;

[0010] The video surveillance node set of each candidate path after removing duplicate identification is obtained by the above step as input, and a multi-point joint identification and detection method is proposed to calculate the minimum set of monitoring nodes that meets the preset requirements;

[0011] Compare the pedestrian feature similarities of the monitoring nodes in each valid alternative path in the previous step, and determine that the suspected target passenger with the greatest feature similarity is the correct value, and determine that the path where the monitoring node with the greatest feature similarity is located is the passenger's actual travel path.

[0012] Preferably, the matching of card swiping data in the automatic ticket vending and checking system, associating the card swiping data, surveillance video and screen position, and achieving binding between multiple categories of information includes:

[0013] Read the card swiping data in the automatic ticket vending and checking system to obtain a data set containing entry and exit information;

[0014] Obtain all the entry gates, exit gates and surveillance video points at the station, mark the position of passengers passing through each gate on the surveillance video screen; and bind gate information with multiple sets of surveillance video and screen position information.

[0015] Preferably, the feature library of pedestrian samples collected from the monitoring video of each card number passing through the entry gate and the exit gate includes:

[0016] Read the card number, entry gate, and entry time information from the card swiping data; find the surveillance video of the entry gate; collect video images of passengers passing through the gate at the time of entry; sample human body features and mark them as entry sample features for the card number;

[0017] Read the card number, exit gate, and exit gate time information from the card swiping data; find the surveillance video of the exit gate; collect the video image of the passenger passing the gate at the exit gate time; sample human body features and mark them as the exit sample features of the card number;

[0018] Identify the complete process of the passenger entering and exiting the station gate from entering the video screen to leaving the video screen; automatically collect all the video image features of the passenger as the sample feature material of the passenger;

[0019] Establish a training sample feature library for the historical travel of pedestrians corresponding to each card number.

[0020] Preferably, the search is for K travel paths of the entry-exit route; invalid paths are eliminated to obtain a valid set of alternative paths, including: when the minimum travel time of the path is greater than or equal to the actual travel time of the passenger and / or the average travel time of the path is greater than M times the average travel time of the shortest path of the OD route, where M is greater than or equal to 1.5, the path is an invalid path.

[0021] Preferably, the steps of calculating a set of surveillance videos of any alternative route passing through each identification station; dividing the itinerary of the alternative route into N segments with the transfer station as a dividing point; and estimating the earliest and latest times of the passenger's appearance at each monitoring point include:

[0022] Based on the distribution map of video surveillance points in the station, a database table of video surveillance points for all passenger walking routes between the starting and ending points in the station and along the way is calculated; based on the starting and ending points of the alternative paths at a certain identification station, the set of surveillance points included in the identification station is searched;

[0023] The alternative route is divided into m segments and m-1 transfers, with the transfer station as the dividing point. The segment range is from the boarding platform to the alighting platform of the route passed by the alternative route, and the transfer starting point and end point are from the alighting platform of the transfer station to the boarding platform of the next route.

[0024] Based on the walking time into the first station, the exit time at the last station, the fixed boarding time for each journey segment, the waiting time, and the walking time for transfers between different journey segments, the earliest and latest times that passengers appear at each video surveillance point are calculated.

[0025] Preferably, the calculation formula for calculating the earliest and latest time that a passenger appears at each video surveillance point is based on the walking time for entering the first station, the exit time for the last station, the fixed boarding time for each journey segment, the waiting time, and the walking time for transfers between different journey segments, including:

[0026]

[0027] Where, t in is the walking time for passengers to enter the station at the starting station O; t out The walking time for passengers to exit the terminal at station D; is the waiting time of the passenger at the starting station of the i-th trip segment; is the fixed running time of the train in the i-th journey segment; is the transfer walking time of passenger i-th transfer.

[0028] Preferably, the calculation of the earliest and latest times when a passenger appears at each video surveillance point based on the walking time into the first station, the exit time from the last station, the fixed boarding time of each journey segment, the waiting time, and the walking time for transfers between different journey segments includes: the calculation formula for the fixed travel time of a passenger on each route segment under path r is:

[0029]

[0030] Where, is the fixed running time of the train in the i-th line segment; t j,stop is the train’s stop time at station j; t j→j+1,run is the travel time of the train from station j to station j+1; is and ie are the starting station and ending station numbers of the i-th journey segment respectively.

[0031] Preferably, the alternative path is divided into m travel segments and m-1 transfers, with the transfer station passing through the alternative path as the dividing point; wherein the travel segment range is from the boarding platform to the getting-off platform of the line through which the alternative path passes, and the transfer starting point-end point is from the getting-off platform of the transfer station to the boarding platform of the next line, including: the complete travel path P of the alternative path r r The expression is:

[0032] p r ={G in ,B r,1 ,(T r,1 ,B r,2 ),…,(T r,i ,B r,i+1 ),…,(T r,m-1 ,B r,m ),G out},m≥1

[0033] T r,i =(S r,i.e ,S r,(i+1).s ),B r,i+1 =(S r,(i+1).s ,S r,(i+1).e ),

[0034] Among them, G in G is the entrance gate of the starting station; out For the exit gate of the terminal station; T r,i is the i-th transfer in the journey of path r, from the alighting station S of the i-th journey segment r,i.e Go to the platform S for the i+1th segment r,(i+1).s ; B r,i+1 is the i+1th segment of path r, including the starting node S of the boarding platform r,(i+1).sTo the end node S of the platform where you get off r,(i+1).e .

[0035] Preferably, the set of video surveillance nodes of each candidate path obtained by the above step after eliminating duplicate identification is used as input, and a scheme for selecting the minimum set of surveillance nodes that meets preset requirements is proposed based on a multi-point joint identification and detection method, including:

[0036] According to the passenger's walking path, the video surveillance nodes of a certain alternative path at each identification station are connected in parallel or in series;

[0037] The detection accuracy of the monitoring nodes is sorted from high to low, and the pedestrian re-identification detection algorithm is called to search for suspected target passengers in the historical videos within the time range corresponding to the monitoring node, and the feature similarity with the target passenger is calculated.

[0038] A subway passenger travel spatiotemporal trajectory recognition system based on pedestrian re-identification is characterized in that it is used to execute any of the above subway passenger travel spatiotemporal trajectory recognition methods based on pedestrian re-identification, including:

[0039] The data access processing module is used to match the card swiping data in the automatic ticket vending and checking system, associate the card swiping data with the surveillance video and the screen position, and realize the binding between multiple categories of information;

[0040] Passenger identity binding module, used to implement identity binding between passenger card numbers and surveillance image features, and to establish a historical image feature library for each passenger;

[0041] The feature sampling modeling module is used to collect video image features of passengers to be compared. It collects the appearance, gait, and motion trajectory features of pedestrians from the surveillance videos at the entry and exit gates where passengers pass through, and builds a deep learning training model for pedestrian re-identification.

[0042] The effective path search module is used to calculate K travel paths for any alternative route, and eliminate invalid paths based on the actual travel time of passengers to obtain effective alternative routes;

[0043] The monitoring point search module is used to search for the set of monitoring nodes of the identification stations passed by each alternative route, calculate the time range of passengers arriving at each monitoring node, and narrow the number and time range of videos to be compared;

[0044] The pedestrian re-identification module is used to search for the target passenger in the historical video data under the constraints of the specified monitoring node and time range, calculate the feature similarity and detection time of the suspected target passenger, and record the results;

[0045] The travel path identification module is used to call the pedestrian re-identification module for the monitoring node of each alternative path, compare the maximum value and occurrence frequency of the feature similarity of each path, and determine the actual path of the passenger.

[0046] The method, system, and assembly method for identifying the spatiotemporal trajectories of subway passengers based on pedestrian re-identification provided by the present invention offer the following advantages: Compared with existing technologies, the method and system for identifying the spatiotemporal trajectories of subway passengers based on pedestrian re-identification establishes a passenger video image feature database for each ticket ID, achieving precise matching between AFC card swipe data and video image data. This method rapidly screens a set of valid alternative travel paths for passengers, focusing on paths requiring comparison and eliminating clearly invalid paths. Surveillance cameras passing along each alternative path are searched to identify the cameras that distinguish different paths, and the time range within which the passenger arrives at each camera is calculated. This reduces the number of cameras to be compared and the time range, reducing ineffective searches. A pedestrian re-identification and comparison scheme based on multi-level camera reconnection is proposed, improving the recognition accuracy of passenger spatiotemporal trajectories and the accuracy of identification hits. A system device for identifying passenger travel trajectories is proposed, automatically performing functions such as feature sampling, model training, feature comparison, and result output, thus achieving automation and intelligence in the recognition system. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 A flowchart of a method for identifying the spatiotemporal trajectories of subway passengers based on pedestrian re-identification provided by an embodiment of the present invention;

[0049] Figure 2 This is the location map of gate 020605 at XJK station (position G5);

[0050] Figure 3 This is the surveillance video and location corresponding to the XJK station entrance gate 020605 (left 1);

[0051] Figure 4 The surveillance video and location corresponding to JMS's exit gate 032202 (2nd from the right);

[0052] Figure 5 There are three alternative routes from XJK Station to JMS Station:

[0053] Figure 6 The second of the three alternative routes from XJK Station to JMS Station;

[0054] Figure 7 There are three alternative routes from XJK station to JMS station;

[0055] Figure 8 This is the distribution map of monitoring points of DXG station;

[0056] Figure 9 Schematic diagram of eliminating duplicate identification stations in two alternative paths Figure 1 ;

[0057] Figure 10 Schematic diagram of eliminating duplicate identification stations for two alternative paths Figure 2 ;

[0058] Figure 11 A schematic diagram of a method for parallel and series cascade combination detection;

[0059] Figure 12 Schematic diagram of the two paths;

[0060] Figure 13 This is a schematic diagram of the escalator layout on platform 0322 of the station;

[0061] Figure 14 This is a schematic diagram of the target passenger 080****5600. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0063] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0064] Currently, all subway stations in cities across China are fully covered by cameras in key areas. For example, in Beijing, approximately 56,000 cameras are installed in 477 stations, with each station boasting over 100 cameras, covering most key areas. This makes it possible to track pedestrian movements using cameras. Therefore, video-based pedestrian re-identification technology can theoretically solve the problem of identifying pedestrian travel paths. However, due to the large number of subway stations across the network (Beijing and Shanghai have over 450 stations) and the large number of cameras at each station, the massive, wide-scale, and long-term spatiotemporal retrieval and comparison of videos requires significant transportation resources and time-consuming computations, which has become a bottleneck in the application and promotion of this technology.

[0065] The following describes in detail the method for identifying the spatiotemporal trajectories of subway passengers based on pedestrian re-identification provided by the embodiments of the present application through specific embodiments and their application scenarios in conjunction with the accompanying drawings.

[0066] Reference Figures 1 to 14 , shows a flowchart of a method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification provided by an embodiment of the present invention. The method may specifically include the following steps:

[0067] Step S1: Match the card swipe data in the Automatic Fare Collection (AFC) system, associate the card swipe data with the surveillance video and the image position, and achieve binding between multiple categories of information. The card swipe data includes entry and exit information.

[0068] Step S1 may include the following sub-steps:

[0069] Sub-step S1.1: Read the AFC card swiping data of the entry and exit, and obtain a data set including the entry information and the exit information.

[0070] Entry information is obtained by reading data from the entry gate. Exit information is obtained by reading data from the exit gate. Entry gate information includes entry records, specifically including: card number, transaction date, entry station, entry time, and entry gate. Exit gate information includes exit records, specifically including: card number, transaction date, exit station, exit time, and exit gate.

[0071] Sub-step S1.2: Obtain all the entry gate groups, exit gate groups and surveillance video points of the station, mark the position of passengers passing through each gate in the surveillance video screen; and realize the binding of gate information with multiple sets of surveillance video and screen position information.

[0072] The station gate group consists of multiple gates, each of which corresponds to a surveillance video. The position where the passenger appears on the surveillance video screen when swiping the card to pass through the gate is marked, and the gate code (including entry gate and exit gate), surveillance video point, and screen position information are bound together.

[0073] Step S2: Collect the pedestrian sample feature library of each card number when passing through the entrance and exit gates. Specifically, read each complete AFC card swiping data and collect the pedestrian sample feature library of each card number when passing through the entrance and exit gates.

[0074] Step S1 ensures that when a passenger swipes a card to pass through the gate in this step, a pedestrian feature image corresponding to the monitoring video point, gate-passing time, and characteristic screen position can be accurately collected.

[0075] Step S2 may include the following sub-steps:

[0076] Sub-step S2.1, read the card number, entry gate, and entry time information of the card swiping data; find the surveillance video of the entry gate; collect video images of passengers passing through the gate at the time of entry in the surveillance video; sample human body features and mark them as entry sample features of the card number.

[0077] Sub-step S2.2, read the card number, exit gate, and exit gate time information of the card swiping data; find the surveillance video of the exit gate; collect the video image of the passenger passing the gate at the exit gate time in the surveillance video; sample human body features and mark them as the exit sample features of the card number.

[0078] Sub-step S2.3: Based on the Yolo-v5 pedestrian detection algorithm and the Deep-sort tracking algorithm, the complete process of the passenger entering and exiting the station gate from entering the video screen to leaving the video screen is identified; all video image features of the passenger are automatically collected as the sample feature material of the passenger.

[0079] Sub-step S2.4: Combine the historical travel records of each card number and establish a training sample feature library for the historical travel of the pedestrian corresponding to each card number.

[0080] If the card number type is not a one-way ticket, the historical travel records of the card number can be retrieved, and the sample features collected historically can be added to the sample feature library of the day for model training.

[0081] The sample features of each pedestrian include pedestrian appearance features, gait features, motion trajectory features, etc.

[0082] In one example, referring to Figure 2and Table 1, wherein Table 1 is the complete AFC card swiping data from XJK station (0206) to JMS station (0322) of a certain city subway line network; Figure 2 The location map of gate 020605 at XJK station is shown.

[0083] Table 1 Complete AFC record from XJK to JMS

[0084]

[0085] Read the first record, determine that the entry station is 0206, the entry gate is 020605, and find the location distribution map corresponding to the gate (refer to Figure 2 ) and find the corresponding video screen of the gate (refer to Figure 3 ).

[0086] The surveillance video of the passenger with card number 080****5600 at XJK station gate 020605 (refer to Figure 3 When there is a time deviation between the AFC system clock and the video surveillance system clock, a time correction operation is performed. After locating the pedestrian, the pedestrian detection and tracking algorithm can be used to calculate the complete video clip of pedestrian 080****5600 passing through the gate and use it as a training model sample.

[0087] The surveillance video of the passenger with card number 080****5600 at JMS exit gate 032202 (refer to Figure 4 After locating the pedestrian, the pedestrian detection and tracking algorithm is used to calculate the complete video clip of pedestrian 080****5600 passing through the gate as a training model sample.

[0088] Step S3: Using the complete AFC card swipe data from step S2 and the information about the entry station O and exit station D, search for K travel paths corresponding to the OD route. Calculate the average travel time, en route time, and the passenger's actual travel time for each path. Eliminate invalid paths to obtain a set of valid alternative paths.

[0089] If any of the following conditions are met in the path searched in step S3, it is considered an invalid path for the passenger:

[0090] ①The minimum travel time of the route is greater than or equal to the actual travel time of the passenger;

[0091] ②The average travel time of this path is greater than M times the average travel time of the shortest path of the OD route, where M is greater than or equal to 1.5.

[0092] It should be noted that in this step, the minimum travel time for each route does not include the waiting time for each journey segment, that is, it is composed of: the minimum walking time to enter the starting station, the fixed train running time for each journey segment, the minimum transfer walking time when passing through the transfer station, and the minimum exit walking time at the terminal station.

[0093] The fixed running time of a train for each journey segment is composed of the running time of the section passed by the journey segment + the stop time at the station passed by the journey segment.

[0094] The passenger's actual travel time = exit gate time - entry gate time.

[0095] The average travel time of this route is composed of: the average walking time into the starting station, the fixed train running time of each journey segment, the average waiting time of each journey segment, the average transfer walking time at the transfer station, and the average exit walking time at the terminal station.

[0096] In one example, a simple subway network consists of four lines. There are three possible routes from XJK station to JMS station, which are: Figure 5 、 Figure 6 、 Figure 7 The calculated in-transit and average travel times for the three routes from XJK Station to JMS Station are shown in Table 2. Based on the arrival and departure times of passenger 080****5600, the actual travel time is calculated to be 1108 seconds. All three routes meet the first invalid path determination rule. According to the second determination rule, with M = 1.5, the shortest path travel time * 1.5 = 1335 seconds. The third route, 1785 seconds > 1335 seconds, is considered invalid. Therefore, after eliminating route 3, the valid routes are routes 1 and 2.

[0097] Table 2 Travel route composition and travel time from XJK station to JMS station

[0098]

[0099] Step S4: Based on the distribution of station surveillance video points, calculate the set of surveillance video clips for each identified station along a candidate route. Divide the route into N segments, using transfer stations as the dividing points. Based on the walking time to the first station, the walking time to the last station, the fixed boarding time for each segment, and the transfer walking time between different segments, calculate the earliest and latest times a passenger will appear at each surveillance point.

[0100] Step S4 may include the following sub-steps:

[0101] Sub-step S4.1: Based on the station video surveillance point distribution map, calculate a database table of video surveillance point sets for all passenger walking routes between the starting and ending points within the station. Based on the starting and ending points of the candidate paths at a specific identification station, search for the set of surveillance points included in that identification station.

[0102] In one example, Table 3 shows a query table of monitoring points of the DXG station. Figure 8 The distribution map of monitoring points of DXG station is shown in Table 3 and Figure 8 You can query the collection and distribution of monitoring points that pedestrians pass through from different starting locations to end locations.

[0103] Table 3 Monitoring point query table from different starting points to end points of DXG station

[0104]

[0105] Sub-step S4.2: Divide the alternative route into m segments and m-1 transfers (m ≥ 1), using the transfer station as the dividing point. A segment ranges from the boarding platform to the alighting platform of the route along the alternative route. The transfer start-end points are from the alighting platform of the transfer station to the boarding platform of the next route.

[0106] In this step, the complete travel path P of the alternative path r r The expression is:

[0107] p r ={G in ,B r,1 ,(T r,1 ,B r,2 ),…,(T r,i ,B r,i+1 ),…,(T r,m-1 ,B r,m ),G out},m≥1

[0108] T r,i =(S r,i.e ,S r,(i+1).s ),B r,i+1 =(S r,(i+1).s ,S r,(i+1).e ),

[0109] Among them, G in G is the entrance gate of the starting station; out For the exit gate of the terminal station; T r,i is the i-th transfer in the journey of path r, from the alighting station S of the i-th journey segment r,i.e Go to the platform S for the i+1th segment r,(i+1).s ; B r,i+1is the i+1th segment of path r, including the starting node S of the boarding platform r,(i+1).s To the end node S of the platform where you get off r,(i+1).e

[0110] The complete surveillance video set C of the alternative path r at all identification stations r , the set C of monitoring points that the path passes through on the walking route into the station, the bus route, the transfer walking route, and the walking route out of the station r,in 、C r,B 、C r,T 、C r,out The union of ; the solution expression is:

[0111] C r =σ in C r,in Uσ B C r,B Uσ T C r,T Uσ out C r,out

[0112] where

[0113] C r,in =f(G in ,S r,1.s ),

[0114] C r,out =f(S r,m.e ,G out ),

[0115]

[0116] Where: C r,in is the camera set on the walking route into the station, f(G in ,S r,1.s,1 ) for passengers entering the station from Gate G in Walk to the starting node S of the first route segment r,1.s,1 Collection of surveillance cameras along the way; C r,out is the camera set on the outbound walking route, f(S r,m.e,n ,G out ) represents the passenger's departure from the end node S of trip segment m r,m.e,n Walk to exit gate G out Collection of surveillance cameras passing by; C r,B is the camera set on the bus route, f(S r,i.s ,S r,i.e ) indicates that the passenger starts from the boarding station of the i-th route segment at node S r,i.s To the end node S of the next platform r,i.eA collection of surveillance cameras passing by, To find the union of m route segments. r,T is the camera set on the transfer route, f(S r,i.e ,S r,(i+1).s ) represents the end node S of the passenger getting off the platform at the i-th line segment r,i.e Transfer to the starting node S of the boarding platform of the i+1th line segment r,(i+1).s A collection of surveillance cameras passing by, To find the union of m-1 route segments. in , σ B , σ T , σ out It is a Boolean variable that determines whether the camera set is selected.

[0117] Furthermore, the function f(x, y) is used to calculate the set of monitoring points that a passenger passes through from the starting node x to the ending node y. This can be obtained by using the table lookup method in sub-step S4.1.

[0118] Furthermore, the priority of selecting the camera set is C r,T >C r,in =C r,out >C r,B ; σ T The default value is 1, B , σ T , σ out The default value is 0; if the first station is a transfer station, σ in =1; if the last station is a transfer station, then σ out =1; if the alternative path contains a loop, to distinguish the different directions of the loop, then σ B =1.

[0119] Sub-step S4.3: Based on the walking time to the first stop, the exit time to the last stop, the fixed boarding time for each journey segment, the waiting time, and the transfer walking time between different journey segments, calculate the earliest and latest time that the passenger appears at each video surveillance point.

[0120] Step S4.3 may include the following sub-steps:

[0121] Substep S4.3.1: For a given OD pair, the total travel time for alternative route r is composed of the walking time to the station, the waiting time for the journey segments passing through all lines, the fixed travel time, the transfer walking time for the transfer journey segments, and the walking time to the station, as shown in the following formula:

[0122]

[0123] Where, t in is the walking time for passengers to enter the station at the starting station O; tout The walking time for passengers to exit the terminal at station D; is the waiting time of the passenger at the starting station of the i-th trip segment; is the fixed running time of the train in the i-th journey segment; is the transfer walking time of the passenger for the i-th transfer; m is the number of journey segments.

[0124] Furthermore, the fixed travel time of each line segment of a passenger's journey under path r is the sum of the stop time at all stations in the direction of travel of the journey segment and the travel time in the forward section;

[0125]

[0126] Where, is the fixed running time of the train in the i-th line segment; t j,stop is the train’s stop time at station j; t j→j+1,run is the travel time of the train from station j to station j+1; is and ie are the starting station and ending station numbers of the i-th journey segment respectively.

[0127] Sub-step S4.3.2: Based on the entry gate time, forwardly calculate the earliest time when the passenger appears at the monitoring point j of each identification station k; based on the exit gate time, reversely calculate the latest time when the passenger appears at the monitoring point j of each identification station k.

[0128] The implementation process of this step can be as follows: First, calculate the earliest time when the passenger appears at the monitoring point j of the identification station k:

[0129]

[0130] Where: T in The time when passengers enter the station and pass through the gate; is the cumulative travel time of the passenger in the first k-1 segments; if k-1≤0, this item is 0; is the cumulative transfer walking time of the passenger at the first k-2 identification stations. The identification stations include: the first station, the transfer stations along the way, and the last station; if k-2≤0, then this item is 0; t in,min is the minimum walking time for passengers entering the first station; σ1 is a Boolean variable; g(pk.s ,p k.j ) is the passenger's starting position p from the identification station k k.s To monitoring point j position p k.j Minimum walking time;

[0131] Then, the latest time when the passenger appears at the identification station in the i-th journey segment is calculated as:

[0132]

[0133] Where: T out The time it takes for passengers to exit the station; is the cumulative travel time of the passenger in the last mk segments; if mk≤0, this item is 0; is the cumulative transfer walking time of the passenger at the next mk-1 transfer stations. If mk-1≤0, then this item is 0; t out,min is the minimum walking time for passengers to enter the terminal station; σ2 is a Boolean variable; g(p k.j ,p k.e ) Passenger from monitoring point j to position p k.j To the end position p of identification station k k.e Minimum walking time;

[0134] Furthermore, the time range of a passenger’s appearance at monitoring point j of each identification station k is [T kj,ET ,T kj,LT ], search duration:

[0135] t kj,find =T kj,LT -T kj,ET

[0136] Furthermore, the acceleration ratio of the actual travel time of the passenger to the search time calculated based on this method is calculated:

[0137]

[0138] From the above formula, we can see that the larger the ratio is, the more time is saved and the higher the calculation efficiency is.

[0139] The use of this step can effectively solve the problems of search range and search efficiency of surveillance cameras and the problem of accurately estimating the search time range for cross-lens pedestrian re-identification.

[0140] In one example, the calculation results of the travel segment division, identification station and walking route, and search time range of the two paths are shown in Table 4 and Table 5:

[0141] The passenger's entire journey took 1108 seconds, and the search time for each node on Path 1 and Path 2 was 418 seconds and 408 seconds, respectively. The calculated speedup ratios were 2.65x and 2.72x, significantly improving the efficiency of time search.

[0142] Table 4 Search time range for each segment of route 1 from XJK station to JMS station

[0143]

[0144] Table 5 Search time range for each segment of route 2 from XJK station to JMS station

[0145]

[0146]

[0147] Step S5: Compare the complete path set of each candidate path in the previous step, remove the overlapping identification stations, and obtain a filtered comparison path set and monitoring video node set.

[0148] In one example, referring to Figures 9 and 10 , taking 2 paths as an example:

[0149] Route 1: 0101G1-0101DN-0102DN / 0408UP-0404UP / 0302DN-0306DN-0306G2 Route 2: 0101G1-0101DN-0105DN / 0208UP-0204UP / 0305DN-0306DN-0306G2 Note: DN stands for downlink platform, UP stands for uplink platform

[0150] Since the starting platform 0101DN and the terminal platform 0306DN of Path 1 and Path 2 are the same, the surveillance video collections in which passengers appear at the first and last stations can be eliminated.

[0151] The paths of Path 1 and Path 2 after elimination are actually the transfer nodes in the middle of the comparison, as shown below:

[0152] Identification stations for path 1: [0102DN / 0408UP], [0404UP / 0302DN]

[0153] Identification stations for path 2: [0105DN / 0208UP], [0204UP / 0305DN]

[0154] Therefore, the surveillance video node sets of path 1 and path 2 after elimination are:

[0155]

[0156]

[0157] Step S6: The set of video surveillance nodes of each alternative path after eliminating duplicate identifications obtained in the above step is used as input. In order to improve the accuracy of pedestrian re-identification on the path, a multi-point joint recognition and detection method is proposed to calculate the minimum set of monitoring nodes that meets the minimum error.

[0158] Step S6 may include the following sub-steps:

[0159] Sub-step S6.1: Connect the video surveillance nodes of a certain alternative path at each identification station in parallel or series according to the passenger's walking path; the parallel-connected monitoring nodes are regarded as a combined node, and its pedestrian re-identification detection accuracy is the product of the detection accuracy of the sub-monitoring nodes.

[0160] Furthermore, if the detection accuracy γ of the monitoring point is an evaluation of the pedestrian re-identification accuracy of the surveillance camera based on a priori samples, γ∈[0,1]. For the parallel structure, the combined detection accuracy is the product of the re-identification accuracy of all cameras at the monitoring node; the expression is as follows:

[0161]

[0162] Where: A r,j,P is the detection accuracy of path r in parallel at the jth monitoring node; γ j,k is the prior detection accuracy of the kth camera under the jth monitoring node, and n is the number of cameras at the video monitoring point.

[0163] For the combined monitoring accuracy of the series structure = 1-the product of the re-identification errors of all cameras under the monitoring node, the expression is:

[0164]

[0165] Where: A r,j,S is the detection accuracy of path r connected in series at the jth monitoring node.

[0166] Furthermore, the parallel relationship can be physically interpreted as that at a certain video monitoring point, due to the large width of the detection area, a single surveillance camera cannot cover it, and two or more cameras are required to detect side by side.

[0167] Sub-step S6.2: Sort the detection accuracy of the monitoring nodes from large to small, call the pedestrian re-identification detection algorithm, search for suspected target passengers in the historical video within the time range corresponding to the monitoring node (obtained by step S4.3.2), and calculate the feature similarity with the target passenger. If the similarity exceeds the set threshold, it is determined that the target passenger is found, and records such as the monitoring point location, similarity value, and appearance time are recorded.

[0168] The detection process of the pedestrian re-identification algorithm includes the following steps:

[0169] 1. In step S2, the pedestrian sample features of the pedestrian i whose card number is to be detected in the entrance gate and exit gate monitoring video are obtained as the sample features of the target to be found. The vector is represented by F i ;

[0170] 2. For the surveillance video to be compared, the earliest and latest times of pedestrian i calculated based on step S4 are used as the search range. YOLO-V5 algorithm is used for multi-target detection, and Deep-sort is used for target tracking. The features corresponding to all pedestrian IDs in this period are calculated and output. The vector is represented as F j , j represents the jth pedestrian in this period.

[0171] 3. Calculate the similarity between the features of the target pedestrian i and all the features of the pedestrian targets j to be compared. The similarity is determined using the Euclidean distance. The formula is as follows, and the feature similarities are sorted from large to small.

[0172]

[0173] x i,s 、x j,s is the s-th sub-feature of the feature pedestrian i, j, and n is the number of features of pedestrian i.

[0174] 4. Take the pedestrian target with the largest feature similarity and determine whether its similarity exceeds the set threshold. If it exceeds, it is determined that the target to be detected has been found.

[0175] To reduce invalid re-identification detections while meeting the expected re-identification accuracy target, the minimum error threshold ε is set and the number of monitoring nodes g that need to be compared is calculated to ensure that the combined detection error E(g) of the first g series nodes is less than the threshold ε. The formula is as follows:

[0176]

[0177] Where, Represents the product value of the identification and detection errors of the first g monitoring points.

[0178] In one example, referring to Figure 11 , assuming that the re-identification detection accuracy of c1, c2, and c3 is 0.9, the re-identification detection accuracy of c4 and c5 is 0.8, and the re-identification detection accuracy of c7, c8, and c9 is 0.9; then:

[0179] The cameras c1, c2, and c3 of the monitoring node M1 are connected in parallel, and λ1 = 1. The detection accuracy at this location is:

[0180]

[0181] Cameras c4 and c5 of monitoring node M2 ​​are connected in series, and λ2 = 0. Then the detection accuracy at this location is:

[0182]

[0183] Here, monitoring node M2 ​​is selected first for re-identification, and monitoring node M1 is considered second:

[0184] Set the minimum convergence error ε = 0.01, then through the series combination of M1 and M2, the combined detection error is E(m) = (1-0.729) × (1-0.96) ≈ 0.01, which has met the convergence condition. If m = 2, stop searching for the subsequent M3 monitoring node.

[0185] Step S7: Compare the pedestrian feature similarities of the monitoring nodes in each valid alternative path in the previous step, and determine that the suspected target passenger with the greatest feature similarity is the correct value, and determine that the path where the monitoring node with the greatest feature similarity is located is the passenger's actual travel path.

[0186] For example, the two paths are shown as Figure 12 On the walking route from the station 0322 up platform to the exit gate 032202 on route 1, the surveillance camera on the west escalator (as shown Figure 13 As shown in ), find the target pedestrian whose feature similarity is the largest. The actual image of the target is found as Figure 14 As shown, it can be inferred that the passenger's actual travel path is path 1.

[0187] This invention provides a method for identifying the spatiotemporal trajectories of subway passengers based on pedestrian re-identification. It innovatively integrates AFC card swiping data and train timetable data, and is based on ReID technology. It solves the following problems: 1. How to achieve identity binding between passenger ticket IDs and pedestrian image features and automatic feature sampling; 2. How to scientifically select monitoring points to be compared and the retrieval time range to improve image comparison and retrieval efficiency; 3. How to solve the problem of low re-identification accuracy of a single monitoring video through a cascade recombination method; and 4. How to identify passengers' travel paths based on pedestrian re-identification results.

[0188] The present invention provides a method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification, establishes a passenger video image feature database for each ticket ID, and realizes accurate matching of AFC card swiping data and video image data. It quickly screens a set of valid alternative travel paths for passengers, focuses on the paths that need to be compared, and eliminates some obviously invalid paths. It searches for surveillance cameras passing through each alternative path, finds surveillance cameras used to distinguish different paths, and calculates the time range for passengers to arrive at each camera. It reduces the number and time range of cameras to be compared, and reduces invalid searches. It proposes a pedestrian re-identification and comparison scheme based on multi-level camera reconnection to improve the recognition accuracy of passengers' spatiotemporal trajectory paths. It improves the hit accuracy of recognition. It proposes a system device for passenger travel trajectory recognition, which automatically realizes the functions of feature sampling, model training, feature comparison, and result output. It realizes the automation and intelligence of the recognition system.

[0189] The present invention also provides a system for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification, which is used to execute the steps in a method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification as described in any one of the above items, including: a data access processing module, a passenger identity binding module, a feature sampling modeling module, an effective path search module, a monitoring point search module, a pedestrian re-identification module, and a travel path identification module. The data access processing module is used to match the card swiping data in the automatic ticket vending system, associate the card swiping data, the monitoring video and the screen position, and realize the binding between multiple categories of information; the passenger identity binding module is used to realize the identity binding of the passenger card number and the monitoring image feature, and establish a historical image feature library for each passenger; the feature sampling modeling module is used to collect the video image features of the passengers to be compared, realize the collection of the appearance features, gait features, and motion trajectory features of the pedestrians in the monitoring videos at the entry and exit gates passed by the passengers, and establish a deep learning training model for pedestrian re-identification; the effective path search module is used to calculate K travel paths for any alternative path, and summarize them. The actual travel time of passengers is combined, invalid paths are eliminated, and valid alternative paths are obtained; the monitoring point search module is used to search the set of monitoring nodes of the identification station passed by each alternative path, calculate the time range of the passenger's arrival at each monitoring node, and narrow the number and time range of videos to be compared; the pedestrian re-identification module is used to search for target passengers in historical video data under the constraints of specified monitoring nodes and time range, calculate the feature similarity and detection time of suspected target passengers, and record the results; the travel path identification module is used to call the pedestrian re-identification module for the monitoring node of each alternative path, compare the maximum value and frequency of feature similarity of each path, and determine the actual path of the passenger.

[0190] As for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0191] An embodiment of the present invention further provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the various processes of the above-mentioned embodiment of the method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification are implemented, and the same technical effects can be achieved. To avoid repetition, these processes are not described here.

[0192] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the various processes of the embodiment of the method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification are implemented, and the same technical effects can be achieved. To avoid repetition, they are not described here.

[0193] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0194] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, apparatus, or computer program products. Thus, embodiments of the present invention may take the form of a fully hardware embodiment, a fully software embodiment, or an embodiment combining software and hardware. Furthermore, embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0195] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0196] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0197] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0198] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic creative concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0199] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element.

[0200] The above describes in detail the method, system, electronic device, and computer-readable storage medium for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification provided by the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will appreciate that, based on the concepts of the present invention, variations may occur in the specific implementation methods and scope of application. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A method for identifying the spatiotemporal trajectories of subway passengers based on pedestrian re-identification, characterized by: The following steps are involved: Match card swiping data in the automatic ticket vending and checking system, associate card swiping data, surveillance video and screen position, and realize the binding between multiple categories of information; Collect a sample feature library of pedestrians with each card number from surveillance videos at the entrance and exit gates; Search for K travel paths for the inbound-outbound route; Eliminate invalid paths and obtain a set of valid alternative paths; Calculate the surveillance video set of any alternative path passing through each identification station; The alternative route is divided into N segments using the transfer station as the dividing point; Calculate the earliest and latest time a passenger appears at each monitoring point; Compare the complete path set of each alternative path in the previous step, remove the overlapping identification stations, and obtain the filtered comparison path set and monitoring video node set; The video surveillance node set of each candidate path after removing duplicate identification is obtained by the above step as input, and a multi-point joint identification and detection method is proposed to calculate the minimum set of monitoring nodes that meets the preset requirements; Compare the pedestrian feature similarities of the monitoring nodes in each valid alternative path in the previous step, and determine that the suspected target passenger with the greatest feature similarity is the correct value, and determine that the path where the monitoring node with the greatest feature similarity is located is the passenger's actual travel path.

2. The method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification according to claim 1, characterized in that: The matching of card swiping data in the automatic ticket vending and checking system, associating the card swiping data, surveillance video and screen position, and realizing the binding between multiple categories of information includes: Read the card swiping data in the automatic ticket vending and checking system to obtain a data set containing entry and exit information; Obtain all the entry gates, exit gates and surveillance video points at the station, mark the position of passengers passing through each gate on the surveillance video screen; and bind gate information with multiple sets of surveillance video and screen position information.

3. The method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification according to claim 2, characterized in that: The pedestrian sample feature library collected from the surveillance video of each card number passing through the entry gate and exit gate includes: Read the card number, entry gate, and entry time information from the card swiping data; find the surveillance video of the entry gate; collect video images of passengers passing through the gate at the time of entry; sample human body features and mark them as entry sample features for the card number; Read the card number, exit gate, and exit gate time information from the card swiping data; find the surveillance video of the exit gate; collect the video image of the passenger passing the gate at the exit gate time; sample human body features and mark them as the exit sample features of the card number; Identify the complete process of the passenger entering and exiting the station gate from entering the video screen to leaving the video screen; automatically collect all the video image features of the passenger as the sample feature material of the passenger; Establish a training sample feature library for the historical travel of pedestrians corresponding to each card number.

4. The method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification according to claim 3, characterized in that: The search is for K travel paths for the inbound-outbound route; Eliminate invalid paths and obtain a set of valid alternative paths, including: when the minimum travel time of the path is greater than or equal to the actual travel time of the passenger and / or the average travel time of the path is greater than M times the average travel time of the shortest path of the route, where M is greater than or equal to 1.5, the path is considered invalid.

5. The method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification according to claim 3, characterized in that: The calculation of any alternative path passing through a set of surveillance videos of each identification station; The alternative route is divided into N segments using the transfer station as the dividing point; Calculate the earliest and latest time a passenger appears at each monitoring point, including: Based on the distribution map of video surveillance points in the station, a database table of video surveillance points for all passenger walking routes between the starting and ending points in the station and along the way is calculated; based on the starting and ending points of the alternative paths at a certain identification station, the set of surveillance points included in the identification station is searched; The alternative route is divided into m segments and m-1 transfers, with the transfer station as the dividing point. The segment range is from the boarding platform to the alighting platform of the route passed by the alternative route, and the transfer starting point and end point are from the alighting platform of the transfer station to the boarding platform of the next route. Based on the walking time into the first station, the exit time at the last station, the fixed boarding time for each journey segment, the waiting time, and the walking time for transfers between different journey segments, the earliest and latest times that passengers appear at each video surveillance point are calculated.

6. The method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification according to claim 5, characterized in that: The calculation formula for the entire travel time is to calculate the earliest and latest time a passenger appears at each video surveillance point based on the walking time to the first stop, the exit time to the last stop, the fixed boarding time for each journey segment, the waiting time, and the walking time for transfers between different journey segments. Where, t in is the walking time for passengers to enter the station at the starting station O; t out The walking time for passengers to exit the terminal at station D; is the waiting time of the passenger at the starting station of the i-th trip segment; is the fixed running time of the train in the i-th journey segment; is the transfer walking time of passenger i-th transfer.

7. The method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification according to claim 6, characterized in that: The calculation formula for calculating the earliest and latest time a passenger appears at each video surveillance point is based on the walking time into the first station, the exit time from the last station, the fixed boarding time for each journey segment, the waiting time, and the walking time for transfers between different journey segments. The calculation formula includes: Where, is the fixed running time of the train in the i-th line segment; t j,stop is the train’s stop time at station j; t j→j+1,run is the travel time of the train from station j to station j+1; is, ie are the starting station and ending station numbers of the i-th journey segment respectively.

8. The method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification according to claim 5, characterized in that: The alternative path is divided into m segments and m-1 transfers, with the transfer station passing through the alternative path as the dividing point; wherein the range of the segment is from the boarding platform to the alighting platform of the line through which the alternative path passes, and the transfer starting point-end point is from the alighting platform of the transfer station to the boarding platform of the next line, including: the complete travel path P of the alternative path r r The expression is: p r ={G in ,B r,1 ,(T r,1 ,B r,2 ),...,(T r,i ,B r,i+1 ),...,(T r,m-1 ,B r,m ),G out },m≥1 T r,i =(S r,i.e ,S r,(i+1).s ),B r,i+1 =(S r,(i+1).s ,S r,(i+1).e ), Among them, G in G is the entrance gate of the starting station; out For the exit gate of the terminal station; T r,i is the i-th transfer in the journey of path r, from the alighting station S of the i-th journey segment r,i.e Go to the platform S for the i+1th segment r,(i+1).s ; B r,i+1 is the i+1th segment of path r, including the starting node S of the boarding platform r,(i+1).s To the end node S of the platform where you get off r,(i+1).e .

9. The method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification according to claim 5, characterized in that: The above step calculates the set of video surveillance nodes of each candidate path after eliminating duplicate identification as input, and proposes a multi-point joint identification and detection method to calculate the minimum set of monitoring nodes that meets the preset requirements, including: According to the passenger's walking path, the video surveillance nodes of a certain alternative path at each identification station are connected in parallel or in series; The detection accuracy of the monitoring nodes is sorted from high to low, and the pedestrian re-identification detection algorithm is called to search for suspected target passengers in the historical videos within the time range corresponding to the monitoring node, and the feature similarity with the target passenger is calculated.

10. A subway passenger travel spatiotemporal trajectory recognition system based on pedestrian re-identification, characterized by: The method for identifying the spatiotemporal trajectory of subway passengers based on pedestrian re-identification according to any one of claims 1 to 9 comprises: The data access processing module is used to match the card swiping data in the automatic ticket vending and checking system, associate the card swiping data with the surveillance video and the screen position, and realize the binding between multiple categories of information; Passenger identity binding module, used to implement identity binding between passenger card numbers and surveillance image features, and to establish a historical image feature library for each passenger; The feature sampling modeling module is used to collect video image features of passengers to be compared. It collects the appearance, gait, and motion trajectory features of pedestrians from the surveillance videos at the entry and exit gates where passengers pass through, and builds a deep learning training model for pedestrian re-identification. The effective path search module is used to calculate K travel paths for any alternative route, and eliminate invalid paths based on the actual travel time of passengers to obtain effective alternative routes; The monitoring point search module is used to search for the set of monitoring nodes of the identification stations passed by each alternative route, calculate the time range of passengers arriving at each monitoring node, and narrow the number and time range of videos to be compared; The pedestrian re-identification module is used to search for the target passenger in the historical video data under the constraints of the specified monitoring node and time range, calculate the feature similarity and detection time of the suspected target passenger, and record the results; The travel path identification module is used to call the pedestrian re-identification module for the monitoring node of each alternative path, compare the maximum value and occurrence frequency of the feature similarity of each path, and determine the actual path of the passenger.

Images