Regional road network vehicle path reconstruction deduction method and system based on multi-source data fusion

Abstract

The invention provides a regional road network vehicle path reconstruction deduction method and system based on multi-source data fusion, and the method comprises the steps: placing collected video monitoring data into a YOLO series or SSD deep learning detection model for training, and obtaining a vehicle detection model; and obtaining a vehicle detection result by adopting the vehicle detection model after iterative optimization. Then, according to the license plate number, the vehicle position and the sequence of the detection time points, a detection node information chain is constructed for the vehicle; a path reconstruction deduction data set is constructed for the detection node information chain, an objective function of the path reconstruction deduction data set is constructed, constraint conditions of the objective function are set, and a path deduction optimization model is obtained; and finally, calculating and analyzing by adopting a breadth-first search algorithm and a path deduction optimization model to obtain an optimal path of the vehicle in the road network. Compared with the prior art, the method provided by the invention solves the problem that a correct path is difficult to find during path reconstruction under a limited camera.

Description

Technical Field

[0001] The present invention belongs to the technical field of traffic data analysis, and in particular relates to a method and system for reconstructing and deducing vehicle paths in a regional road network based on multi-source data fusion. Background Art

[0002] As the core link of data-driven, road network path reconstruction can accurately capture and reconstruct the dynamic changes of the actual road network, provide real-time and accurate traffic flow information for the digital twin model, improve the virtual-real mapping accuracy of the vehicle path information of the traffic digital twin, enhance the simulation accuracy of the digital twin, and enable it to simulate more complex traffic scenarios, thereby helping traffic management departments to make accurate decisions and efficient scheduling. Therefore, road network path reconstruction is of great significance to the construction of high-precision and high-practicality road network digital twins.

[0003] In the prior art, there are various methods for realizing road network path reconstruction, mainly including the following:

[0004] 1) Methods based on GPS positioning data, such as interpolation, particle filtering and Kalman filtering, predict and supplement the vehicle's location information by processing GPS data to achieve trajectory reconstruction. Such methods may be affected by the quality of GPS signals, such as signal shielding, multipath effects, etc., resulting in limited positioning accuracy. In addition, only a few vehicles have GPS data, and it is difficult to accurately describe all vehicles on the road network using this data.

[0005] 2) Methods based on wireless signals, such as Bluetooth and WiFi signals, infer the location and trajectory of the vehicle by sensing the strength and interaction of the signal. The disadvantages of this type of method are that the signal transmission process is easily interfered by environmental noise, and the signal coverage is limited, which may affect the accuracy and continuity of trajectory reconstruction.

[0006] 3) The video surveillance-based method uses video image processing and computer vision technology to detect and track vehicles and obtain their driving trajectories. Although this method can overcome the problems of the above methods (1) and (2), it has high requirements for image clarity and high costs, and it is difficult to cover regional road networks. Summary of the invention

[0007] The present invention provides a method and system for reconstructing and deducing vehicle paths in a regional road network based on multi-source data fusion, which are used to solve the problems existing in the prior art.

[0008] In a first aspect, to achieve the above-mentioned purpose, the present invention provides a method for reconstructing and deriving a vehicle path in a regional road network based on multi-source data fusion, comprising:

[0009] S10: For the road network in the target area, the video surveillance data collected within the selected monitoring time is placed into the YOLO series or SSD deep learning detection model for training to obtain a vehicle detection model;

[0010] S20: Divide the video surveillance data into a training data set and a detection data set, place the training data set into a vehicle detection model for iterative training, and place the detection data set into the iteratively trained vehicle detection model to obtain vehicle detection results in the video surveillance data;

[0011] The vehicle detection result includes at least the license plate number, the detection time point and the vehicle position corresponding to the detection time point;

[0012] S30: Building a detection node information chain for the vehicle to be tested according to the license plate number and the vehicle location and in the order of detection time points;

[0013] S40: using the reconstruction deduction method to construct a path reconstruction deduction data set for the detection node information chain;

[0014] S50: constructing an objective function of the path reconstruction deduction data set, and setting constraints of the objective function to obtain a path deduction optimization model;

[0015] S60: Using the breadth-first search algorithm and path deduction optimization model, the optimal path of the vehicle to be tested in the road network is calculated and analyzed.

[0016] Through the above technical scheme, the method of the present invention can integrate the multimodal information features of machine vision and vehicle positioning data under the condition of limited number of cameras and limited clarity, construct the camera node path of each vehicle based on the time chain, build a path deduction optimization model based on the fusion of machine vision and vehicle positioning data, and use the breadth-first search algorithm and the path deduction optimization model to complete the deduction of the optimal path for the vehicle.

[0017] Preferably, the vehicle detection result also includes vehicle volume, vehicle color and vehicle shape, and step S20 includes:

[0018] S21: Using the video surveillance data as a sample data set, dividing the sample data set into a training data set and a detection data set, and performing annotation and augmentation processing;

[0019] S22: placing the training data set into the vehicle detection model for iterative training;

[0020] S23: placing the detection data set into the iteratively trained vehicle detection model, scanning each frame in the video surveillance video, identifying and marking the vehicle position, vehicle volume, license plate number, detection time point, vehicle color and vehicle shape;

[0021] S24: Use the target tracking algorithm to continuously track the vehicle, predict the position of the vehicle in the future frame, and match the information with the current vehicle detection result.

[0022] Preferably, the target tracking algorithm adopts Kalman filtering or DeepSORT algorithm.

[0023] Preferably, after step S23, the method of the present invention further comprises:

[0024] Calculate the vehicle's speed, specifically:

[0025] S200: Acquire the positions of multiple vehicles within a selected time range;

[0026] S201: using a road network matching algorithm, simulating the positions of multiple vehicles into multiple vehicle trajectories and matching them to road sections of the road network;

[0027] S203: selecting a statistical time interval of a time range, and dividing the plurality of vehicle trajectories into a plurality of corresponding statistical data sets according to the statistical time interval;

[0028] S204: Calculate the average driving speed of vehicles in each road section according to the statistical data set.

[0029] Preferably, the objective function in step S50 is:

[0030]

[0031] in, TR is the time required for the vehicle to travel from the starting point to the end point. start,end The detection time length TI of the surveillance camera start，end The difference between

[0032] The constraints in step S50 include at least:

[0033] |TR start,end -TI start,end |≤μ

[0034] Among them, μ is the allowable error threshold.

[0035] Preferably, step S60 includes:

[0036] S61: Setting the starting node and the end point of the road network topology;

[0037] S62: Create a priority search queue and add the start node to the priority search queue;

[0038] S63: Creating a candidate set of reachable paths that meet the constraint conditions;

[0039] S64: adopting a breadth-first search algorithm, cyclically calculating the time-consuming costs of candidate paths in the reachable path candidate set, and taking the reachable candidate path with the lowest time-consuming cost as the optimal path.

[0040] Preferably, the breadth-first search algorithm of the method of the present invention is specifically:

[0041] S641: Take out and remove the node from the head of the priority search queue, and record the taken out node as the current node;

[0042] S642: Obtaining a set of adjacent nodes according to the current node and the road network topology map;

[0043] S643: Retrieve a candidate path set whose last node is the current node from the candidate path set that meets the constraint condition;

[0044] S644: cyclically calculating the time costs of the candidate paths in the candidate path set corresponding to the current node, and taking the reachable candidate path with the lowest time cost as the optimal path.

[0045] Through the above technical solution, the method of the present invention includes an improvement on the breadth-first search algorithm. In order to avoid the process of calculating the reachable path using the breadth-first search algorithm from falling into an infinite loop of a cyclic graph, the present invention avoids the process of searching for a path from falling into an infinite loop through steps S641-S644.

[0046] Preferably, before step S10, the method of the present invention further comprises:

[0047] S100: Select monitoring time range;

[0048] S101: Filter out raw video surveillance data that falls within a surveillance time range;

[0049] S102: Preprocess the original video surveillance data to obtain video surveillance data.

[0050] In a second aspect, in order to achieve the above-mentioned purpose, the present invention further provides a regional road network vehicle path reconstruction deduction system based on multi-source data fusion, using the regional road network vehicle path reconstruction deduction method based on multi-source data fusion, the system of the present invention comprises:

[0051] The vehicle detection model is used to train the video surveillance data collected within the selected monitoring time into the YOLO series or SSD deep learning detection model for the target area road network to obtain the vehicle detection model;

[0052] The detection result acquisition module is used to divide the video surveillance data into a training data set and a detection data set, place the training data set into the vehicle detection model for iterative training, and place the detection data set into the iteratively trained vehicle detection model to obtain the vehicle detection results in the video surveillance data;

[0053] The vehicle detection result includes at least the license plate number, the detection time point and the vehicle position corresponding to the detection time point;

[0054] The node information chain building module is used to build a detection node information chain for the vehicle to be tested according to the license plate number and vehicle location and the order of detection time points;

[0055] A data set construction module is used to construct a path reconstruction and deduction data set for the detection node information chain by using a reconstruction and deduction method;

[0056] The path deduction optimization module is used to construct the objective function of the path reconstruction deduction data set and set up the constraint conditions of the objective function to obtain the path deduction optimization model;

[0057] The optimal path acquisition module is used to use the breadth-first search algorithm and the path deduction optimization model to calculate and analyze the optimal path for the vehicle in the road network;

[0058] Among them, the objective function of the path deduction optimization module is:

[0059]

[0060] in, TR is the time required for the vehicle to travel from the starting point to the end point. start，end The detection time length TI of the surveillance camera start，end The difference between

[0061] The constraints in step S50 include at least:

[0062] |TR start，end -TI start，end |≤μ

[0063] Among them, μ is the allowable error threshold.

[0064] The method and system for regional road network vehicle path reconstruction and deduction based on multi-source data fusion of the present invention have at least the following advantages:

[0065] 1) Compared with the prior art, the method of the present invention solves the problem of finding the correct path when reconstructing the path under limited cameras, and can find the most likely driving path of the vehicle with the smallest error with the actual time.

[0066] 2) Compared with the prior art, the method of the present invention proposes a regional road network vehicle path reconstruction and deduction model that integrates vehicle positioning data and machine vision. The method takes into account the vehicle running speed at different time points on different road sections, and finds the most likely vehicle driving path through a breadth-first search algorithm.

[0067] 3) Compared with the prior art, the present invention considers vehicle information such as the license plate number and monitoring time point recognized by each camera in the road network based on the vehicle detection model, filters the detection information of the corresponding vehicle according to the license plate number, and sorts it from front to back according to the monitoring time point to build a detection node information chain. The detection node information chain is also divided according to adjacent nodes. The obtained vehicle path reconstruction and deduction data set is more in line with the actual situation and can improve the accuracy of the final output result.

[0068] 4) Compared with the prior art, the method of the present invention includes an improvement on the traditional breadth-first search algorithm. In order to avoid the process of calculating the reachable path using the breadth-first search algorithm from falling into an infinite loop of a cyclic graph, the present invention avoids the process of searching for a path from falling into an infinite loop through steps S641-S644. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0070] Figure 1 This is a flow chart of a method for reconstructing and deducing a vehicle path in a regional road network based on multi-source data fusion in Embodiment 1 of the present invention.

[0071] Figure 2 It is a road network diagram of the method for reconstructing and deducing a vehicle path in a regional road network based on multi-source data fusion in Example 1 of the present invention.

[0072] Figure 3 It is the optimal path diagram of the regional road network vehicle path reconstruction and deduction method based on multi-source data fusion in Example 1 of the present invention.

[0073] Figure 4 It is a structural schematic diagram of a regional road network vehicle path reconstruction and deduction system based on multi-source data fusion in Example 2 of the present invention. DETAILED DESCRIPTION

[0074] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0075] In the description of the present invention, it should be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they cannot be understood as limitations on the present invention.

[0076] This embodiment discloses a method and system for reconstructing and deducing vehicle paths in a regional road network based on multi-source data fusion, which are used to solve the problems existing in the prior art.

[0077] Embodiment 1

[0078] like Figure 1 As shown, this embodiment discloses a method for reconstructing and deducing a vehicle path in a regional road network based on multi-source data fusion, including:

[0079] S10: For the road network in the target area, the video surveillance data collected within the selected monitoring time is placed into the YOLO series or SSD deep learning detection model for training to obtain a vehicle detection model.

[0080] In this embodiment, if Figure 2 As shown, the above video surveillance data is obtained through a series of data processing by cameras set up in the road network of the target area.

[0081] Specifically, before step S10, the method of the present invention further includes:

[0082] S100: Select monitoring time range;

[0083] S101: Filter out raw video surveillance data that falls within a surveillance time range;

[0084] S102: Preprocess the original video surveillance data to obtain video surveillance data.

[0085] It should be specifically explained that, in this embodiment, the starting time of the above monitoring time range is recorded as DT st , and the end time is recorded as DT edThe raw video surveillance data is obtained by cameras installed on the road network. The raw video surveillance data includes the road section monitoring data and the camera’s own information, such as camera location information, serial number information, tilt degree, monitoring range, etc.

[0086] Furthermore, the above step S101 is specifically as follows: firstly, the original video surveillance data is used as the camera information set VD = {vd1, vd2, ..., vd z} (z represents the camera number), and then calibrate the detection range and tilt degree of each camera, and set the starting time of the subsequent model sample data as TDT st , the end time is recorded as TDT ed , and then filter the subsequent model training and evaluation data sets, recorded as Data_Sample, and finally according to the DT st To DT ed The video surveillance raw data of each camera is filtered within the time range of , and the data is used as the actual data for path reconstruction and recognition, recorded as Data_Real.

[0087] Preferably, the above preprocessing includes denoising, contrast enhancement and frame extraction.

[0088] S20: Divide the video surveillance data into a training data set and a detection data set, place the training data set into a vehicle detection model for iterative training, and place the detection data set into the iteratively trained vehicle detection model to obtain vehicle detection results in the video surveillance data.

[0089] In this embodiment, the vehicle detection result includes vehicle position, vehicle volume, license plate number, detection time point, vehicle color, vehicle shape, vehicle paint texture, vehicle speed and vehicle acceleration.

[0090] Specifically, the above step S20 includes:

[0091] S21: Using the video surveillance data as a sample data set, dividing the sample data set into a training data set and a detection data set, and performing annotation and augmentation processing;

[0092] S22: placing the training data set into the vehicle detection model for iterative training;

[0093] S23: placing the detection data set into the iteratively trained vehicle detection model, scanning each frame in the video surveillance video, identifying and marking the vehicle position, vehicle volume, license plate number, detection time point, vehicle color and vehicle shape;

[0094] S24: Use the target tracking algorithm to continuously track the vehicle, predict the position of the vehicle in the future frame, and match the information with the current vehicle detection result.

[0095] Compared with the prior art, the method of the present invention first divides the sample data set Data_Sample into a training data set and a detection data set, and performs annotation and augmentation processing, and then uses the training data set to iteratively train the vehicle detection model to improve the accuracy of detection. The iterative vehicle detection model needs to be evaluated until it meets the expected effect. After the vehicle detection model training is completed, the detection data set Data_Real is placed in the model, and each frame in the video surveillance video is scanned to realize the feature extraction of the vehicle. Finally, the Kalman filter or DeepSORT algorithm is used to continuously track the detected vehicle, predict the position of the vehicle in the future frame, and match the information with the actual detected vehicle, so as to expand the vehicle trajectory of each vehicle in the road network.

[0096] It should be explained that the method of the present invention can further calculate the vehicle's driving speed based on the vehicle detection result. This step is performed after the above step S23, and is specifically:

[0097] S200: Acquire the positions of multiple vehicles within a selected time range;

[0098] S201: using a road network matching algorithm, simulating the positions of multiple vehicles into multiple vehicle trajectories and matching them to road sections of the road network;

[0099] S203: selecting a statistical time interval of a time range, and dividing the plurality of vehicle trajectories into a plurality of corresponding statistical data sets according to the statistical time interval;

[0100] S204: Calculate the average driving speed of vehicles in each road section according to the statistical data set.

[0101] The above step S204 includes:

[0102]

[0103] Among them, the first vehicle trajectory point PS of the vehicle entering the road section start and the last trajectory point with non-zero velocity PS end The GIS distance of the road section is denoted as d r ;PS start The corresponding start time point is recorded as RST start , P.S. end The corresponding end time point is recorded as RST end .v * is the average speed of vehicles on each road section within the statistical time interval, e is the total number of vehicles after screening, and v ... k is the speed of the kth vehicle. The speed v of each vehicle on a certain road section can be calculated by formula (1), and the average speed of the vehicles can be obtained by formula (2).

[0104] It should be explained that due to the presence of roadside parking in the road network, if the average speed of each calculated vehicle is added to the road section, the accuracy of the average speed of the road section will be reduced. Therefore, it is necessary to set filtering conditions to eliminate abnormal situations, filter samples, and eliminate floating vehicle data that do not meet the conditions. The last trajectory point is the last trajectory point with a non-zero speed, which has eliminated the impact of vehicle parking at the intersection. Therefore, it can be filtered by the vehicle travel speed threshold condition, and the speed threshold is set to v threshold , the travel speed of each vehicle is related to v threshold Compare, if it is less than v threshold Then remove the travel speed of the vehicle if it is greater than v threshold The final screening result is recorded as RV = {v1, v2, ..., v e}, e is the serial number of the vehicle after screening.

[0105] Compared with the prior art, the average vehicle speed of a road section calculated by the method of the present invention is more in line with the actual situation and can better assist in simulating or predicting the vehicle's driving trajectory.

[0106] After the above step S20, the method of the present invention further comprises:

[0107] S30: Building a detection node information chain for the vehicle to be detected according to the license plate number and the vehicle position and in the order of detection time points.

[0108] Specifically, the vehicle detection model can obtain the spatiotemporal distribution and license plate number of vehicles within the calibration range of each camera, and sort the detection information of each vehicle from front to back according to the license plate number and the order of detection time points. Then, taking the camera as the detection node, a detection node information chain is constructed for each vehicle according to the time sequence of the vehicle detection results. The node information chain of a single vehicle is recorded as Link = {I1, I2, ..., I y}, y is the camera number that the vehicle passes through in the road network driving path, and the vehicle detection node information includes the license plate number, the time point when the vehicle is within the calibration range, the road section position of the vehicle within the calibration range, and the running speed of the vehicle within the calibration range.

[0109] S40: Using the reconstruction deduction method, a path reconstruction deduction data set is constructed for the detection node information chain.

[0110] In this embodiment, the reconstruction deduction method is a deduction method commonly used in the prior art to deduce a vehicle driving path.

[0111] Specifically, after obtaining the detection node information chain of each vehicle, due to the limited number of cameras and the limitations of the camera detection range, the vehicle driving path between adjacent nodes of the detection node information chain is still unknown. Therefore, it is necessary to reconstruct and deduce the vehicle driving path between adjacent nodes of the detection node information chain and construct a path reconstruction deduction dataset for the detection node information chain of each vehicle.

[0112] Furthermore, the link composed of the adjacent nodes of the node information link Link of a single vehicle is used as the deduction data set, which is recorded as the set TLink = {[I1, I2], [I2, I3], [I y-1 , I y ]}, each element in the set is a data set that needs to be reconstructed and deduced.

[0113] It is necessary to further explain that, since vehicles may pass through intersections during driving, there may or may not be traffic lights at the intersections, and the time that vehicles need to wait for traffic lights when they arrive at the intersections is also different, and it is also possible that vehicles pass through intersections but do not need to wait, such as the right turn phase. Therefore, the traffic light waiting time corresponding to different time points when the vehicle arrives at the intersection is also different. The present invention takes into account the traffic light waiting time of vehicles passing through intersections when constructing the path reconstruction deduction, so it is necessary to calculate the traffic light time of each phase of all intersections in the regional road network.

[0114] Specifically, after the above step S40, the method of the present invention further needs to obtain the traffic light information of each phase at each intersection.

[0115] S50: construct an objective function of the path reconstruction deduction data set, and set constraints of the objective function to obtain a path deduction optimization model.

[0116] Specifically, the above objective function is:

[0117]

[0118] in, TR is the time required for the vehicle to travel from the starting point to the end point. start，end The detection time length TI of the surveillance camera start，end The difference.

[0119] The constraints in step S50 at least include:

[0120] |TR start，end -TI start，end |≤μ (4)

[0121] Among them, μ is the allowable error threshold.

[0122] Furthermore, the above step S50 is from each vehicle's TLink={[I1,I2],[I2,I3],[I y-1 ,I y ]} Take out each path deduction data set in order, and the starting point and end point of each path deduction data set are represented by I start ,I end Represents. Since the actual time spent by vehicles traveling at adjacent vehicle path nodes can be obtained through the vehicle detection model, it is known. Therefore, the goal of the path deduction optimization model is to find a reachable path so that the time spent on the path is the smallest difference from the time spent on the adjacent vehicle path nodes, so formula (3) is constructed as its objective function. When the time spent on the reachable path from the start point to the end point is much greater than or much less than the time length detected by the surveillance camera, it means that the path is not the actual driving path of the vehicle. Considering that the average driving speed is used to calculate the path time, it may cause certain errors. Therefore, the difference between the time spent on the reachable path from the start point to the end point and the time length detected by the surveillance camera is set within a certain range as a constraint condition.

[0123] S60: Using the breadth-first search algorithm and path deduction optimization model, the optimal path of the vehicle to be tested in the road network is calculated and analyzed.

[0124] Specifically, the above step S60 includes:

[0125] S61: Set the starting node and end point of the road network topology.

[0126] In this embodiment, the network topology starting nodes V s and the end point V d , the time point corresponding to camera detection is recorded as VT s and VT d , and then calculate the starting node V s To the end point V d The length of the time interval is calculated as follows:

[0127] VTL s，d =VT d -VT s (5)

[0128] Among them, VTL s，d The vehicle starts from the topological starting node V s To the end point V s The actual time interval.

[0129] S62: Create a priority search queue E a , and the starting node V s Add to the priority search queue.

[0130] S63: Create a reachable path candidate set and a reachable path set that meet the constraint conditions.

[0131] In this embodiment, the reachable path candidate set is recorded as CRC={CR 1 ,...,CR m}, after creating the CRC, it is initialized to contain only the starting node V s The constructed path is added to the reachable path candidate set CRC. Among them, m is the number of candidate paths, each element in the set is a candidate reachable path that meets the constraints, each node in the path is the endpoint node of the road segment in the road network topology, and each node contains the time point and time length from the starting point to the current node in the path.

[0132] In this embodiment, the reachable path candidate set is recorded as ARC={AR 1 ,...,AR d}, d is the starting node V s and the end point V d The number of reachable paths.

[0133] S64: adopting a breadth-first search algorithm, cyclically calculating the time-consuming costs of candidate paths in the reachable path candidate set, and taking the reachable candidate path with the lowest time-consuming cost as the optimal path.

[0134] It should be emphasized that the above breadth-first search algorithm is an improved breadth-first search algorithm, which has a built-in ring detection algorithm.

[0135] Furthermore, the breadth-first search algorithm of the present invention is specifically:

[0136] S641: Search from priority queue E a The first node of the team is taken out and removed, and the node taken out is recorded as the current node V current .

[0137] S642: According to the current node V current Get the adjacent node set from the road network topology, denoted as VNC current .

[0138] S643: Retrieve the candidate path set that meets the constraint condition and the last node of the path is V current The candidate path set is denoted as CRC * .

[0139] S644: Circular calculation of the current node V current Corresponding candidate path set CRC * The time cost of the candidate paths is calculated, and the reachable candidate path with the lowest time cost is taken as the optimal path.

[0140] Wherein, step S644 includes:

[0141] S6441: Record the candidate path of the current loop as CR i .

[0142] S6442: Circularly search the current node V in the road network topology map current Neighbor Node Sets and VNC current , specifically:

[0143] Record the adjacent node currently visited as V neighbor , perform repeated node detection on vehicle paths.

[0144] It should be explained that in order to avoid the process of calculating the reachable path using the breadth-first search algorithm from falling into an infinite loop of the cyclic graph, the present invention uses repeated node detection, that is, a ring detection algorithm, to avoid the process of searching for the path from falling into an infinite loop. neighbor The candidate path CR currently visited i If there are repeated nodes in the path, it means that the path contains a cycle graph, then return to step S6442; if there are no repeated nodes, then proceed to the next step.

[0145] S6443: Set the adjacent node V neighbir With candidate path CR i Splice to get new candidate paths

[0146] S6444: Calculate the current node V current Go to adjacent node V neighbor The time you need to wait at the traffic light.

[0147] It needs to be explained that Figure 3 As shown in the figure, since the starting point of the link is the location of the camera, there is no traffic light between the starting point and the first intersection, and there may not be traffic lights between intersections. At the same time, the intersection phase corresponding to the path between adjacent nodes may not have a traffic light waiting time, such as the right turn phase. Therefore, it is necessary to determine whether there is a traffic light waiting time between two adjacent nodes. Determine the candidate path based on the road network topology and traffic light phase Current node V current To its adjacent node V neighbor Do you need to wait for the traffic light?

[0148] For the current node V current To its adjacent node V neighbor The path between the two nodes has no traffic light waiting time, such as the current node V current V s or when there are no traffic lights between adjacent intersections.

[0149] WLn current =0 (6)

[0150] For the case where there is a waiting time for traffic lights between adjacent nodes, the time VT at which the vehicle arrives at the current node is calculated. current And V calculated from the vehicle positioning trajectory data current The node corresponds to the green light time point at the intersection, recorded as GVT current , the calculation formula is as follows:

[0151] WLn current =GVT current -VT current (7)

[0152] Among them, WVTL current is the current node V current The time required to wait for the traffic light to reach its adjacent node.

[0153] S6445: Calculate the vehicle passing through the current node V current To its adjacent node V neighbor The time between the sections.

[0154] Record the current node V in the network topology current To its adjacent node V neighbor The road section between curreny,neighbor , the corresponding section length is RL current,neighbor The time when the vehicle enters the road section is GVT current , calculate according to the vehicle positioning data, and obtain the road section P current,neighbor In time GVT current The average speed of vehicles at point , denoted as SP current,neighbor , the calculation formula is as follows:

[0155]

[0156] Among them, RVTL current The vehicle passes through road section P current,neighbor The length of time it takes.

[0157] S6446: Calculate the vehicle from the current node V current To its adjacent node V neighbor Time spent.

[0158] Note that the vehicle starts from the current node V current To its adjacent node V neighbor Time spent on VTL current,neighbor The length of time includes the waiting time at the traffic light and the time it takes for the vehicle to pass through the road section. The calculation formula is as follows:

[0159] VTLcurrent，neighbor =WVTL current +RVTL current，neighbor (9)

[0160] S6447: Calculate candidate paths The vehicle starts from the starting point V s To the adjacent node V neighbor The time point is calculated as follows:

[0161] VT neighbor =VT current +VTL current，neighbor (10)

[0162] If the current node V current Starting point V s , then VT current VT s .

[0163] S6448: Calculate candidate paths The vehicle starts from the starting point V s To Node V neighbor time cost.

[0164] It should be explained that the present invention takes into account the different time points when vehicles pass through different sections and the different average speeds of the sections, and proposes a time cost calculation method based on the time-varying characteristics of the link. The cost is calculated from the starting point V s To Node V neighbor The time cost includes the vehicle's driving time at different time periods on different sections of the road and the waiting time at traffic lights at different time points when passing through the intersection. The calculation formula is as follows:

[0165] VTL s，neighbor =VT neighbor -VT s (11)

[0166] S6449: Determine the time cost VTL s,neighbor Whether the constraints of the vehicle path reconstruction simulation model are met.

[0167] Calculate candidate paths The vehicle starts from the starting point V s To Node V neighbor Time cost of VTL s,neighbor The actual time interval VTL from the start point to the end point s,d The error is calculated as follows:

[0168] ε=||VTL s，d |-VTL s，neighbor | (12)

[0169] If ε is greater than the given error μ, it means that the path does not meet the constraint conditions. If the search continues, the error will be larger, thus reducing unnecessary searches for the path and returning to step B to continue visiting other adjacent nodes. If it is less than or equal to the given error μ, then the path meets the constraint conditions and proceeds to the next step.

[0170] After the above step S6449, determine the candidate path for splicing The last node V neighbor Is it the end point V d If yes, then it means that a reachable path that meets the constraints from the starting point to the end point has been found, and the path is added to the reachable path set ARC. If no, then the path is a candidate path that meets the constraints and can be added to the candidate path set CRC. The last node V neighbor Join the priority search queue a , and finally the candidate path CR i Delete CRC from the candidate path set. Then, determine E a Is it empty? If it is empty, the loop ends, that is, all reachable nodes have been visited. Finally, according to the objective function, find the reachable path with the smallest error ε in the set of reachable paths from the starting point to the end point ARC. Recorded as CR ′ .

[0171] Compared with the prior art, the method of steps S61-S64 can obtain the optimal path of the adjacent node information chain of each vehicle, and splice the optimal paths deduced for each pair of adjacent nodes to construct a complete optimal driving path for the vehicle in the road network.

[0172] Through the above steps S10-S60, compared with the prior art, the method of the present invention has the following advantages:

[0173] (1) The present invention can combine video monitoring data and vehicle positioning data under limited camera conditions, integrate the multimodal information features of machine vision and vehicle positioning data, construct the camera node path of each vehicle based on the time chain, and consider the dynamic running speed of vehicles and the waiting time of traffic lights in different time periods of the road section. A path reconstruction and deduction model integrating machine vision and vehicle positioning data is constructed to complete the reconstruction and deduction of vehicle driving paths at the road network level.

[0174] (2) The present invention addresses the problem of finding the correct path when reconstructing a path under limited camera conditions. An improved breadth-first search algorithm is used to find the most likely vehicle travel path with the smallest error from the actual time.

[0175] Embodiment 2

[0176] like Figure 4 As shown, this embodiment discloses a regional road network vehicle path reconstruction and deduction system based on multi-source data fusion, which adopts the regional road network vehicle path reconstruction and deduction method based on multi-source data fusion in embodiment 1. The system of the present invention includes:

[0177] Vehicle detection model, which is used to put the collected video surveillance data into the YOLO series or SSD deep learning detection model for training;

[0178] A detection result acquisition module is used to obtain vehicle detection results in the monitoring image by using a target tracking algorithm and an iteratively optimized vehicle detection model;

[0179] The vehicle detection result includes at least the license plate number, the detection time point and the vehicle position corresponding to the detection time point;

[0180] The node information chain building module is used to build a detection node information chain for the vehicle according to the license plate number and vehicle location, in the order of detection time points;

[0181] The data set construction module is used to construct a path reconstruction deduction data set for the detection node information chain;

[0182] The path deduction optimization module is used to construct the objective function of the path reconstruction deduction data set and set up the constraint conditions of the objective function to obtain the path deduction optimization model;

[0183] The optimal path acquisition module is used to use the breadth-first search algorithm and the path deduction optimization model to calculate and analyze the optimal path for the vehicle in the road network.

[0184] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, etc. Various media that can store program codes.

[0185] For those skilled in the art, various other corresponding changes and deformations can be made according to the technical solutions and concepts described above, and all of these changes and deformations should fall within the protection scope of the claims of the present invention.

Claims

1. A vehicle path reconstruction and deduction method for a regional road network based on multi-source data fusion, characterized in that: include: S10: For the road network in the target area, the video surveillance data collected within the selected monitoring time is placed into the YOLO series or SSD deep learning detection model for training to obtain a vehicle detection model; S20: Divide the video surveillance data into a training data set and a detection data set, place the training data set into the vehicle detection model for iterative training, and place the detection data set into the iteratively trained vehicle detection model to obtain vehicle detection results in the video surveillance data; The vehicle detection result includes at least the license plate number, the detection time point and the vehicle position corresponding to the detection time point; S30: Building a detection node information chain for the vehicle to be detected according to the license plate number and the vehicle position and in the order of the detection time points; S40: Using a preset reconstruction deduction method to construct a path reconstruction deduction data set for the detection node information chain; S50: constructing an objective function of the path reconstruction deduction data set, and setting constraints of the objective function to obtain a path deduction optimization model; S60: using a breadth-first search algorithm and the path deduction optimization model to calculate and analyze to obtain an optimal path for the vehicle to be tested in the road network; Wherein, the objective function in step S50 is: in, TR is the time required for the vehicle to travel from the starting point to the end point. start，end The detection time length TI of the surveillance camera start，end The difference between The constraint conditions in step S50 at least include: |TR start，end -YOU start，end ||≤μ Among them, μ is the allowable error threshold.

2. The method for reconstructing and deducing a regional road network vehicle path based on multi-source data fusion according to claim 1 is characterized in that: The vehicle detection result also includes vehicle volume, vehicle color and vehicle shape. Step S20 includes: S21: taking the video surveillance data as a sample data set, dividing the sample data set into a training data set and a detection data set, and performing annotation and augmentation processing; S22: placing the training data set into the vehicle detection model for iterative training; S23: placing the detection data set into the iteratively trained vehicle detection model, scanning each frame in the video surveillance video, identifying and marking the vehicle position, the vehicle volume, the license plate number, the detection time point, the vehicle color and the vehicle shape; S24: Continuously track the vehicle using a target tracking algorithm, predict the position of the vehicle in future frames, and match the information with the current vehicle detection result.

3. The method for reconstructing and deducing a regional road network vehicle path based on multi-source data fusion according to claim 2 is characterized in that: The target tracking algorithm adopts Kalman filtering or DeepSORT algorithm.

4. The method for reconstructing and deducing a regional road network vehicle path based on multi-source data fusion according to claim 2 is characterized in that: After step S23, the method further includes: Calculate the driving speed of the vehicle, specifically: S200: Acquire the positions of multiple vehicles within a selected time range; S201: using a road network matching algorithm, simulating the positions of multiple vehicles into multiple vehicle trajectories and matching them to road sections of the road network; S203: selecting a statistical time interval of the time range, and dividing the plurality of vehicle trajectories into a plurality of corresponding statistical data sets according to the statistical time interval; S204: Calculate the average driving speed of vehicles in each road section according to the statistical data set.

5. The method for reconstructing and deducing a regional road network vehicle path based on multi-source data fusion according to claim 1, characterized in that: The step S60 comprises: S61: Setting the starting node and the end point of the road network topology; S62: Create a priority search queue, and add the start node to the priority search queue; S63: Creating a reachable path candidate set and a reachable path set that meet the constraint conditions; S64: adopting a breadth-first search algorithm to cyclically calculate the time-consuming costs of the candidate paths in the reachable path candidate set, and taking the reachable candidate path with the lowest time-consuming cost as the optimal path.

6. The method for reconstructing and deducing a regional road network vehicle path based on multi-source data fusion according to claim 5 is characterized in that: The breadth-first search algorithm is specifically: S641: taking out and removing a node from the head of the priority search queue, and recording the taken out node as the current node; S642: Acquire a set of adjacent nodes according to the current node and the road network topology map; S643: Retrieve a candidate path set whose last node in the path is the current node from the candidate path set that meets the constraint condition; S644: cyclically calculating the time costs of the candidate paths in the candidate path set corresponding to the current node, and taking the reachable candidate path with the lowest time cost as the optimal path.

7. The method for reconstructing and deducing a regional road network vehicle path based on multi-source data fusion according to claim 1, characterized in that: Before step S10, the method further includes: S100: Select monitoring time range; S101: Filter out raw video surveillance data that falls within the monitoring time range; S102: Preprocessing the original video surveillance data to obtain the video surveillance data.

8. Regional road network vehicle path reconstruction and deduction system based on multi-source data fusion, characterized by: The method for reconstructing and deducing a regional road network vehicle path based on multi-source data fusion according to any one of claims 1 to 7 is adopted, and the system comprises: The vehicle detection model is used to train the video surveillance data collected within the selected monitoring time into the YOLO series or SSD deep learning detection model for the target area road network to obtain the vehicle detection model; A detection result acquisition module, used for dividing the video surveillance data into a training data set and a detection data set, placing the training data set into the vehicle detection model for iterative training, and placing the detection data set into the iteratively trained vehicle detection model to obtain vehicle detection results in the video surveillance data; The vehicle detection result includes at least the license plate number, the detection time point and the vehicle position corresponding to the detection time point; A node information chain building module is used to build a detection node information chain for the vehicle to be tested according to the license plate number and the vehicle position and in the order of the detection time points; A data set construction module, used to use a reconstruction deduction method to construct a path for the detection node information chain to reconstruct the deduction data set; A path deduction optimization module is used to construct an objective function of the path reconstruction deduction data set and set constraints of the objective function to obtain a path deduction optimization model; An optimal path acquisition module is used to use a breadth-first search algorithm and the path deduction optimization model to calculate and analyze the optimal path of the vehicle to be tested in the road network; Among them, the objective function of the path deduction optimization module is: in, TR is the time required for the vehicle to travel from the starting point to the end point. start，end The detection time length TI of the surveillance camera start，end The difference between The constraint conditions in step S50 at least include: |TR start，end -YOU start，end |≤μ Among them, μ is the allowable error threshold.

Images