Vehicle tracking method and system based on intelligent road nail array

Abstract

The invention provides a vehicle tracking method based on an intelligent road nail array. The method comprises the following steps that intelligent road nail units are arranged along lane marks to obtain a vehicle position signal sequence; feature vectors in the vehicle position signal sequence are automatically extracted according to the de-noising auto-encoder model; a dynamic time warping algorithm is adopted to calculate the similarity of the two feature vectors, and the similarity is utilized to quantify the correlation weight of a plurality of vehicle position signal sequences collected by the intelligent road nail units of the adjacent cross sections; performing optimal correlation on the vehicle position signal sequences of the adjacent cross sections according to a Kuhn-Munkres algorithm to realize matching of the two position signal sequences of the same vehicle on the adjacent cross sections so as to form a vehicle track; high-precision detection and tracking of vehicles are realized by utilizing a magnetic sensor array embedded with lane marks and combining a de-noising auto-encoder model, a dynamic time warping algorithm and a Kuhn-Munkres algorithm; according to the vehicle tracking method based on the intelligent road nail array, the accuracy and robustness of vehicle detection are improved, and the construction and maintenance cost of the system is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of road slope collapse perception detection, and in particular to a vehicle tracking method and system based on an intelligent road spike array. Background Art

[0002] Existing vehicle detection technologies, such as video surveillance systems and radar systems, can meet the needs to a certain extent, but they each have their own limitations. Video surveillance systems are easily affected by lighting and weather conditions, and have large data processing volumes and high computing resource consumption. Although radar systems have strong environmental adaptability, they are expensive and may be interfered by reflections from surrounding objects. Although electromagnetic induction loop detectors have good environmental adaptability, their installation requires excavation of the road surface, which has high construction costs and inconvenient maintenance.

[0003] The new trend of technological development points to the application of multi-sensor fusion technology, intelligent algorithms, and the use of low-power and wireless communication technologies. The development of these technologies makes it possible to overcome the limitations of single sensors, improve the accuracy and robustness of detection systems, and reduce energy consumption and construction costs. Summary of the invention

[0004] The main purpose of the present invention is to overcome the above-mentioned defects in the prior art and propose a vehicle tracking system and method based on an intelligent road stud array. The system uses a magnetic sensor array embedded in lane markings, combined with a denoising autoencoder model, a dynamic time warping algorithm and a Kuhn-Munkres algorithm to achieve high-precision detection and tracking of vehicles. This method not only improves the accuracy and robustness of vehicle detection, but also reduces the construction and maintenance costs of the system.

[0005] The present invention adopts the following technical solution:

[0006] A vehicle tracking method based on an intelligent road stud array comprises the following steps:

[0007] The vehicle position signal sequence is obtained by arranging intelligent road stud units along the lane markings;

[0008] Automatically extract feature vectors from vehicle position signal sequences based on a denoising autoencoder model;

[0009] The dynamic time warping algorithm is used to calculate the similarity of two feature vectors, and the similarity is used to quantify the association weights of multiple vehicle position signal sequences collected by intelligent road stud units in adjacent cross sections;

[0010] According to the Kuhn-Munkres algorithm, the vehicle position signal sequences of adjacent cross sections are optimally associated to achieve the matching of two position signal sequences of the same vehicle in adjacent cross sections to form the vehicle trajectory.

[0011] Specifically, the vehicle position signal is obtained by arranging intelligent road stud units along the lane markings, including identifying interference signals, specifically:

[0012] The method to identify interference signals is:

[0013]

[0014] Where A is the signal strength, A ad represents the signal strength detected by adjacent sensors on the same cross section; f th represents the threshold value, when f is less than f th When IN is 1, it indicates that the signal is an interference signal; otherwise, it is not an interference signal.

[0015] Specifically, the feature vector in the vehicle position signal is automatically extracted according to the denoising autoencoder model; wherein the denoising autoencoder model is specifically:

[0016] The denoising autoencoder model includes an encoder and a decoder, where the encoder includes a one-dimensional convolutional neural network and a fully connected layer;

[0017] The decoder consists of a fully connected layer and a one-dimensional transposed convolutional neural network.

[0018] Specifically, the feature vector in the vehicle position signal is automatically extracted according to the denoising autoencoder model, specifically:

[0019] Vehicle position signal sequence preprocessing:

[0020] The original vehicle position signal sequence X is converted into a new sequence X with a length of 100 by using the linear interpolation method. 100 .

[0021] Feature extraction:

[0022] A one-dimensional convolutional neural network is used to extract local features of the standardized waveform sequence to obtain vector Y;

[0023] Y = Conv1d(X 100 )

[0024] Then, the vector Y is fused and reduced in dimension through the fully connected layer to obtain the feature vector F with a length of the set value S. S ,

[0025] F S =FC(Y).

[0026] The dynamic time warping algorithm is used to calculate the similarity of the two feature vectors, and the similarity is used to quantify the association weights of multiple vehicle position signal sequences collected by the intelligent road stud units in adjacent cross sections, specifically:

[0027] Initialization: Create a two-dimensional dynamic programming matrix M to store local distances. The size of M is equal to the length of the two feature sequences. Create a matrix C of the same shape to store the minimum cumulative distance.

[0028] Calculate the local distance: For two feature sequences in the feature vector F a and F b For each pair of data points in , the local distance is calculated using the Euclidean distance:

[0029] M(i,j)=||F a (i)-F b (j)|| 1≤i,j≤S

[0030] Among them, F a (i) is the feature sequence F a The i-th data point, F b (j) is the feature sequence F b The jth data point; M(i,j) corresponds to the feature sequence F a The i-th data point and the feature sequence F b The local distance of the jth data point;

[0031] Dynamic programming filling: Starting from the upper left corner of the matrix M, calculate the minimum cumulative distance C for each position i,j , which is the sum of the current local distance and the minimum of the left, top, and top-left adjacent elements:

[0032] C i,j =min{C i-1,j-1 ,C i-1,j ,C i,j-1}+M(i,j)

[0033] Returns the minimum cumulative distance: The element in the lower right corner of the matrix represents the minimum cumulative distance, denoted as DTW(F a ,F b ), represents the optimal registration distance between two sequences, where the smaller the optimal registration distance, the higher the similarity.

[0034] Specifically, the vehicle position signal sequences of adjacent cross sections are optimally associated according to the Kuhn-Munkres algorithm to achieve matching of two position signal sequences of the same vehicle in adjacent cross sections to form a vehicle trajectory, wherein the adjacent cross sections are specifically determined as follows:

[0035] Time and space constraints: The time difference between vehicles passing through adjacent sections should satisfy:

[0036]

[0037] Among them, t i and t jrepresents the time when the vehicle passes through two consecutive cross sections, Δt low and Δt up are the lower and upper limits of the time difference;

[0038] Spatial constraint: The number of lanes that a vehicle crosses between two adjacent sections should satisfy:

[0039]

[0040] Among them, l i and l j Represents the identification number of the lane where the two vehicle position signal sequences are located, Δl up Represents the maximum number of lane changes allowed.

[0041]

[0042] Specifically, the vehicle position signal sequences of adjacent cross sections are optimally associated according to the Kuhn-Munkres algorithm to achieve the matching of two position signal sequences of the same vehicle in adjacent cross sections to form the vehicle trajectory, which is specifically:

[0043] Weight w i,j Reflecting the vehicle position signal sequence o i and the vehicle position signal sequence d j The association between them, where the larger the weight, the higher the probability of association between them;

[0044] The inverse of the minimum cumulative distance is used as the weight, so o i and d j The weight w between i,j :

[0045]

[0046] Weight matrix W:

[0047]

[0048] Finally, the Kuhn-Munkres algorithm is used to determine the maximum weight matching set R:

[0049]

[0050] The optimal association between vehicle position signal sequences is achieved, and the matching of two position signal sequences of the same vehicle in adjacent cross sections is achieved.

[0051] Another aspect of the present invention provides a vehicle tracking system based on an intelligent road stud array, specifically comprising:

[0052] Vehicle position signal sequence acquisition unit: acquires the vehicle position signal sequence by using intelligent road stud units arranged along lane markings;

[0053] Feature vector extraction unit: automatically extracts feature vectors from the vehicle position signal sequence based on the denoising autoencoder model;

[0054] Similarity calculation unit: a dynamic time warping algorithm is used to calculate the similarity of two feature vectors, and the similarity is used to quantify the association weights of multiple vehicle position signal sequences collected by the intelligent road stud units in adjacent cross sections;

[0055] Vehicle position signal sequence association unit: According to the Kuhn-Munkres algorithm, the vehicle position signal sequences of adjacent cross sections are optimally associated to achieve the matching of two position signal sequences of the same vehicle in adjacent cross sections to form a vehicle trajectory.

[0056] Another aspect of the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a vehicle tracking method based on an intelligent road stud array is implemented.

[0057] Yet another aspect of the present invention provides a computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, steps of a vehicle tracking method based on an intelligent road stud array are implemented.

[0058] It can be seen from the above description of the present invention that, compared with the prior art, the present invention has the following beneficial effects:

[0059] The present invention proposes a vehicle tracking method based on an intelligent road stud array, which uses intelligent road stud units arranged along lane markings to obtain a vehicle position signal sequence; automatically extracts feature vectors in the vehicle position signal sequence according to a denoising autoencoder model; uses a dynamic time warping algorithm to calculate the similarity of two feature vectors, and uses the similarity to quantify the association weights of multiple vehicle position signal sequences collected by the intelligent road stud units in adjacent cross sections; optimally associates the vehicle position signal sequences of adjacent cross sections according to the Kuhn-Munkres algorithm, achieves matching of two position signal sequences of the same vehicle in adjacent cross sections, and forms a vehicle trajectory; The present invention proposes a vehicle tracking method based on an intelligent road stud array, which uses a magnetic sensor array embedded in lane markings, combined with a denoising autoencoder model, a dynamic time warping algorithm, and a Kuhn-Munkres algorithm to achieve high-precision detection and tracking of vehicles. This method not only improves the accuracy and robustness of vehicle detection, but also reduces the construction and maintenance costs of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 The schematic diagram of the layout of smart road studs (SRS) in the actual road environment;

[0061] Figure 2 A flow chart of a vehicle tracking method based on an intelligent road stud array provided in an embodiment of the present invention;

[0062] Figure 3 A structural diagram of a vehicle tracking method based on an intelligent road stud array provided in an embodiment of the present invention;

[0063] Figure 4 A structural diagram of a denoising autoencoder (DAE) model provided in an embodiment of the present invention;

[0064] Figure 5 Another vehicle tracking system framework diagram based on an intelligent road stud array provided by an embodiment of the present invention;

[0065] Figure 6 A schematic diagram of an electronic device according to an embodiment of the present invention;

[0066] Figure 7 A schematic diagram of an embodiment of a computer-readable storage medium provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The present invention is further described below through specific implementation modes.

[0068] The present invention proposes a vehicle tracking method based on an intelligent road stud array, which uses a magnetic sensor array embedded in lane markings, combined with a denoising autoencoder model, a dynamic time warping algorithm, and a Kuhn-Munkres algorithm to achieve high-precision vehicle detection and tracking. This method not only improves the accuracy and robustness of vehicle detection, but also reduces the construction and maintenance costs of the system.

[0069] like Figure 1 , showing in detail the placement of smart road studs (SRS) in an actual road environment. The location of the SRS is clearly marked in the figure, including their specific layout along the lane markings. Each SRS unit is represented by a diagram, with the distance between them and their specific position relative to the lane marked. In addition, the diagram may also include the angle relationship between the SRS and the lane markings, as well as the monitoring area covered by the SRS array.

[0070] like Figure 2 FIG. 1 is a flow chart of a vehicle tracking method based on an intelligent road stud array according to the present invention; Figure 3 , is a structural diagram of a vehicle tracking method based on an intelligent road stud array, which specifically includes the following steps:

[0071] A vehicle tracking method based on an intelligent road stud array comprises the following steps:

[0072] S201: Acquire a vehicle position signal sequence by using intelligent road stud units arranged along lane markings;

[0073] Obtain the position of the vehicle at different times for subsequent vehicle trajectory tracking. The detection process is mainly divided into the identification of interference signals and the confirmation of vehicle position.

[0074] Generation and identification of interference signals:

[0075] When a vehicle passes through a lane, the detection nodes on both sides of the lane marking will detect magnetic field interference signals. If the intensity of the magnetic field interference signal generated by the vehicle is too high, the detection nodes on non-adjacent lane markings may also detect magnetic field interference signals. These signals are called interference signals.

[0076] The method to identify interference signals is:

[0077]

[0078] Where A is the signal strength, A ad represents the signal strength detected by adjacent sensors on the same cross section; f th represents the threshold value, when f is less than f th When IN is 1, it indicates that the signal is an interference signal; otherwise, it is not an interference signal.

[0079] In addition, special scene processing:

[0080] When the number of lanes exceeds 2, if there is no vehicle passing through the middle lane, but all four detectors detect non-interference signals, it is necessary to determine the vehicle's lane position based on the horizontal and vertical components of the signal.

[0081] S202: Automatically extracting feature vectors from the vehicle position signal sequence according to a denoising autoencoder model;

[0082] The denoising autoencoder model (DAE) is specifically:

[0083] The denoising autoencoder model includes an encoder and a decoder, where the encoder includes a one-dimensional convolutional neural network and a fully connected layer;

[0084] The decoder consists of a fully connected layer and a one-dimensional transposed convolutional neural network. Figure 4 This is the structure diagram of the denoising autoencoder (DAE) model.

[0085] The eigenvectors in the vehicle position signal are automatically extracted according to the denoising autoencoder model, specifically:

[0086] Vehicle position signal sequence preprocessing:

[0087] The original vehicle position signal sequence X is converted into a new sequence X with a length of 100 by using the linear interpolation method. 100 .

[0088] Feature extraction:

[0089] A one-dimensional convolutional neural network is used to extract local features of the standardized waveform sequence to obtain vector Y;

[0090] Y = Conv1d(X 100 )

[0091] Then, the vector Y is fused and reduced in dimension through the fully connected layer to obtain the feature vector F with a length of the set value S. S ,

[0092] F S =FC(Y).

[0093] S203: using a dynamic time warping algorithm to calculate the similarity of the two feature vectors, and using the similarity to quantify the association weights of multiple vehicle position signal sequences collected by the intelligent road stud units in adjacent cross sections;

[0094] The purpose of quantifying the similarity of feature sequences is: Since the feature vector extracted by the one-dimensional convolutional neural network contains time information and the order of elements represents the time series, the dynamic time warping (DTW) algorithm is needed to analyze and quantify the similarity of these feature sequences. Specifically:

[0095] Initialization: Create a two-dimensional dynamic programming matrix M to store local distances. The size of M is equal to the length of the two feature sequences. Create a matrix C of the same shape to store the minimum cumulative distance.

[0096] Calculate the local distance: For two feature sequences in the feature vector F a and F b For each pair of data points in , the local distance is calculated using the Euclidean distance:

[0097] M(i,j)=||F a (i)-F b (j)|| 1≤i,j≤S

[0098] Among them, F a (i) is the feature sequence F a The i-th data point, F b (j) is the feature sequence F b The jth data point; M(i,j) corresponds to the feature sequence F a The i-th data point and the feature sequence F b The local distance of the jth data point;

[0099] Dynamic programming filling: Starting from the upper left corner of the matrix M, calculate the minimum cumulative distance C for each position i,j , which is the sum of the current local distance and the minimum of the left, top, and top-left adjacent elements:

[0100] C i,j =min{C i-1,j-1 ,C i-1,j ,C i,j-1}+M(i,j)

[0101] Returns the minimum cumulative distance: The element in the lower right corner of the matrix represents the minimum cumulative distance, denoted as DTW(F a ,F b ), represents the optimal registration distance between two sequences, where the smaller the optimal registration distance, the higher the similarity.

[0102] Advantages of the DTW algorithm: First, the DTW algorithm effectively measures the similarity of feature vectors containing time information by finding the best time alignment between two time series. Second, this method is particularly suitable for dealing with speed changes and time scale changes, ensuring that signals belonging to the same vehicle can be accurately identified even when the vehicle speed changes or the detection time points are not synchronized.

[0103] S204: optimally correlating the vehicle position signal sequences of adjacent cross sections according to the Kuhn-Munkres algorithm, matching two position signal sequences of the same vehicle in adjacent cross sections, and forming a vehicle trajectory.

[0104] The purpose of multi-target vehicle association matching is to associate multiple vehicle position signal sequences detected by adjacent cross-section magnetic sensors to form vehicle trajectories.

[0105] Solve the optimal matching problem: Solve the optimal matching problem in a bipartite graph and use the Kuhn-Munkres (KM) algorithm to find the maximum weighted matching set.

[0106] Bipartite graph representation: Assume a bipartite graph G = (O, D, E, W).

[0107] The vertex set O represents the vehicle position signal sequence detected in the front section.

[0108] The vertex set D represents the vehicle position signal sequence detected in the rear section.

[0109] The edge set E represents the correspondence between the vehicle position signal sequences in the front and rear sections. i Can satisfy the arrival vehicle position signal sequence d j The time and space constraints of the edge e ij=1, otherwise e ij =0.

[0110] Time and space constraints: The time difference between vehicles passing through adjacent sections should satisfy:

[0111]

[0112] Among them, t i and t j represents the time when the vehicle passes through two consecutive cross sections, Δt low and Δt up are the lower and upper limits of the time difference;

[0113] Spatial constraint: The number of lanes that a vehicle crosses between two adjacent sections should satisfy:

[0114]

[0115] Among them, l i and l j Represents the identification number of the lane where the two vehicle position signal sequences are located, Δl up Represents the maximum number of lane changes allowed.

[0116]

[0117] The weight set W represents the degree of association between signals. i,j Reflecting the vehicle position signal sequence o i and the vehicle position signal sequence d j The larger the weight, the higher the probability of their association, and the greater the possibility that they belong to the same vehicle.

[0118] Edge weight calculation: The inverse of the minimum cumulative distance is used as the weight, so o i and d j The weight w between i,j :

[0119]

[0120] Weight matrix W:

[0121]

[0122] Finally, the Kuhn-Munkre algorithm is used to determine the maximum weight matching set R:

[0123]

[0124] The optimal association between vehicle position signal sequences is achieved, and the matching of two position signal sequences of the same vehicle in adjacent cross sections is achieved.

[0125] like Figure 5 The embodiment of the present invention further provides a vehicle tracking system 500 based on an intelligent road stud array, comprising:

[0126] The vehicle position signal sequence acquisition unit 501 acquires the vehicle position signal sequence by using intelligent road stud units arranged along the lane markings;

[0127] Obtain the position of the vehicle at different times for subsequent vehicle trajectory tracking. The detection process is mainly divided into the identification of interference signals and the confirmation of vehicle position.

[0128] Generation and identification of interference signals:

[0129] When a vehicle passes through a lane, the detection nodes on both sides of the lane marking will detect magnetic field interference signals. If the intensity of the magnetic field interference signal generated by the vehicle is too high, the detection nodes on non-adjacent lane markings may also detect magnetic field interference signals. These signals are called interference signals.

[0130] The method to identify interference signals is:

[0131]

[0132] Where A is the signal strength, A ad represents the signal strength detected by adjacent sensors on the same cross section; f th represents the threshold value, when f is less than f th When IN is 1, it indicates that the signal is an interference signal; otherwise, it is not an interference signal.

[0133] In addition, special scene processing:

[0134] When the number of lanes exceeds 2, if there is no vehicle passing through the middle lane, but all four detectors detect non-interference signals, it is necessary to determine the vehicle's lane position based on the horizontal and vertical components of the signal.

[0135] Feature vector extraction unit 502: automatically extracts feature vectors from the vehicle position signal sequence according to the denoising autoencoder model;

[0136] The denoising autoencoder model (DAE) is specifically:

[0137] The denoising autoencoder model includes an encoder and a decoder, where the encoder includes a one-dimensional convolutional neural network and a fully connected layer;

[0138] The decoder consists of a fully connected layer and a one-dimensional transposed convolutional neural network.

[0139] The eigenvectors in the vehicle position signal are automatically extracted according to the denoising autoencoder model, specifically:

[0140] Vehicle position signal sequence preprocessing:

[0141] The original vehicle position signal sequence X is converted into a new sequence X with a length of 100 by using the linear interpolation method. 100 .

[0142] Feature extraction:

[0143] A one-dimensional convolutional neural network is used to extract local features of the standardized waveform sequence to obtain vector Y;

[0144] Y = Conv1d(X 100 )

[0145] Then, the vector Y is fused and reduced in dimension through the fully connected layer to obtain the feature vector F with a length of the set value S. S ,

[0146] F S =FC(Y).

[0147] Similarity calculation unit 503: using a dynamic time warping algorithm to calculate the similarity of the two feature vectors, and using the similarity to quantify the association weights of multiple vehicle position signal sequences collected by the intelligent road stud units of adjacent cross sections;

[0148] The purpose of quantifying the similarity of feature sequences is: Since the feature vector extracted by the one-dimensional convolutional neural network contains time information and the order of elements represents the time series, the dynamic time warping (DTW) algorithm is needed to analyze and quantify the similarity of these feature sequences. Specifically:

[0149] Initialization: Create a two-dimensional dynamic programming matrix M to store local distances. The size of M is equal to the length of the two feature sequences. Create a matrix C of the same shape to store the minimum cumulative distance.

[0150] Calculate the local distance: For two feature sequences in the feature vector F a and F b For each pair of data points in , the local distance is calculated using the Euclidean distance:

[0151] M(i,j)=||F a (i)-F b (j)|| 1≤i,j≤S

[0152] Among them, F a (i) is the feature sequence F a The i-th data point, F b (j) is the feature sequence F b The jth data point; M(i,j) corresponds to the feature sequence F a The i-th data point and the feature sequence Fb The local distance of the jth data point;

[0153] Dynamic programming filling: Starting from the upper left corner of the matrix M, calculate the minimum cumulative distance C for each position i,j , which is the sum of the current local distance and the minimum of the left, top, and top-left adjacent elements:

[0154] C i,j =min{C i-1,j-1 ,C i-1,j ,C i,j-1}+M(i,j)

[0155] Returns the minimum cumulative distance: The element in the lower right corner of the matrix represents the minimum cumulative distance, denoted as DTW(F a ,F b ), represents the optimal registration distance between two sequences, where the smaller the optimal registration distance, the higher the similarity.

[0156] Advantages of the DTW algorithm: First, the DTW algorithm effectively measures the similarity of feature vectors containing time information by finding the best time alignment between two time series. Second, this method is particularly suitable for dealing with speed changes and time scale changes, ensuring that signals belonging to the same vehicle can be accurately identified even when the vehicle speed changes or the detection time points are not synchronized.

[0157] The vehicle position signal sequence association unit 504 performs optimal association on the vehicle position signal sequences of adjacent cross sections according to the Kuhn-Munkres algorithm, so as to achieve matching of two position signal sequences of the same vehicle in adjacent cross sections and form a vehicle track.

[0158] The purpose of multi-target vehicle association matching is to associate multiple vehicle position signal sequences detected by adjacent cross-section magnetic sensors to form vehicle trajectories.

[0159] Solve the optimal matching problem: Solve the optimal matching problem in a bipartite graph and use the Kuhn-Munkres (KM) algorithm to find the maximum weighted matching set.

[0160] Bipartite graph representation: Bipartite graph representation: Assume a bipartite graph G = (O, D, E, W).

[0161] The vertex set O represents the vehicle position signal sequence detected in the front section.

[0162] The vertex set D represents the vehicle position signal sequence detected in the rear section.

[0163] The edge set E represents the correspondence between the vehicle position signal sequences in the front and rear sections. iCan satisfy the arrival vehicle position signal sequence d j The time and space constraints of the edge e ij =1, otherwise e ij =0.

[0164] Time and space constraints: The time difference between vehicles passing through adjacent sections should satisfy:

[0165]

[0166] Among them, t i and t j represents the time when the vehicle passes through two consecutive cross sections, Δt low and Δt up are the lower and upper limits of the time difference;

[0167] Spatial constraint: The number of lanes that a vehicle crosses between two adjacent sections should satisfy:

[0168]

[0169] Among them, l i and l j Represents the identification number of the lane where the two vehicle position signal sequences are located, Δl up Represents the maximum number of lane changes allowed.

[0170]

[0171] The weight set W represents the degree of association between signals. i,j Reflecting the vehicle position signal sequence o i and the vehicle position signal sequence d j The larger the weight, the higher the probability of their association, and the greater the possibility that they belong to the same vehicle.

[0172] Edge weight calculation: The inverse of the minimum cumulative distance is used as the weight, so o i and d j The weight w between i,j :

[0173]

[0174] Weight matrix W:

[0175]

[0176] Finally, the Kuhn-Munkres algorithm is used to determine the maximum weight matching set R:

[0177]

[0178] The optimal association between vehicle position signal sequences is achieved, and the matching of two position signal sequences of the same vehicle in adjacent cross sections is achieved.

[0179] like Figure 6 As shown, an embodiment of the present invention provides an electronic device 600, including a memory 610, a processor 620, and a computer program 611 stored in the memory 610 and executable on the processor 620. When the processor 620 executes the computer program 611, a vehicle tracking method based on an intelligent road stud array provided in an embodiment of the present invention is implemented.

[0180] In a specific implementation process, when the processor 620 executes the computer program 611, it can achieve Figure 2 Any implementation manner in the corresponding embodiments.

[0181] Since the electronic device introduced in this embodiment is a device used to implement a data processing device in the embodiment of the present invention, based on the method introduced in the embodiment of the present invention, technical personnel in the field can understand the specific implementation of the electronic device of the present embodiment and its various variations. Therefore, how the electronic device implements the method in the embodiment of the present invention is not introduced in detail here. As long as the equipment used by technical personnel in the field to implement the method in the embodiment of the present invention falls within the scope of protection of the present invention.

[0182] See also Figure 7 , Figure 7 A schematic diagram of an embodiment of a computer-readable storage medium provided in an embodiment of the present invention.

[0183] like Figure 7 As shown, this embodiment provides a computer-readable storage medium 700, on which a computer program 711 is stored. When the computer program 711 is executed by a processor, a vehicle tracking method based on an intelligent road stud array provided in an embodiment of the present invention is implemented;

[0184] In the specific implementation process, when the computer program 711 is executed by the processor, it can achieve Figure 2 Any implementation manner in the corresponding embodiments.

[0185] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0186] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, 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 disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0187] The present invention proposes a vehicle tracking method based on an intelligent road stud array, which uses intelligent road stud units arranged along lane markings to obtain a vehicle position signal sequence; automatically extracts feature vectors in the vehicle position signal sequence according to a denoising autoencoder model; uses a dynamic time warping algorithm to calculate the similarity of two feature vectors, and uses the similarity to quantify the association weights of multiple vehicle position signal sequences collected by the intelligent road stud units in adjacent cross sections; optimally associates the vehicle position signal sequences of adjacent cross sections according to the Kuhn-Munkres algorithm, achieves matching of two position signal sequences of the same vehicle in adjacent cross sections, and forms a vehicle trajectory; The present invention proposes a vehicle tracking method based on an intelligent road stud array, which uses a magnetic sensor array embedded in lane markings, combined with a denoising autoencoder model, a dynamic time warping algorithm, and a Kuhn-Munkres algorithm to achieve high-precision detection and tracking of vehicles. This method not only improves the accuracy and robustness of vehicle detection, but also reduces the construction and maintenance costs of the system.

[0188] The above is only a specific implementation of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial changes to the present invention using this concept shall be deemed as an infringement of the protection scope of the present invention.

Claims

1. A vehicle tracking method based on an intelligent road stud array, characterized in that: The steps include: The vehicle position signal sequence is obtained by arranging intelligent road stud units along the lane markings; Automatically extract feature vectors from vehicle position signal sequences based on a denoising autoencoder model; The dynamic time warping algorithm is used to calculate the similarity of two feature vectors, and the similarity is used to quantify the association weights of multiple vehicle position signal sequences collected by intelligent road stud units in adjacent cross sections; According to the Kuhn-Munkres algorithm, the vehicle position signal sequences of adjacent cross sections are optimally associated to achieve the matching of two position signal sequences of the same vehicle in adjacent cross sections to form the vehicle trajectory.

2. A vehicle tracking method based on an intelligent road stud array according to claim 1, characterized in that: The vehicle position signal is obtained by arranging intelligent road stud units along the lane markings, including identifying interference signals, specifically: The method to identify interference signals is: Where A is the signal strength, A ad represents the signal strength detected by adjacent sensors on the same cross section; f th represents the threshold value, when f is less than f th When IN is 1, it indicates that the signal is an interference signal; otherwise, it is not an interference signal.

3. The vehicle tracking method based on the intelligent road stud array according to claim 1, characterized in that: Automatically extract feature vectors from vehicle position signals based on a denoising autoencoder model; The denoising autoencoder model is specifically: The denoising autoencoder model includes an encoder and a decoder, where the encoder includes a one-dimensional convolutional neural network and a fully connected layer; The decoder consists of a fully connected layer and a one-dimensional transposed convolutional neural network.

4. A vehicle tracking method based on an intelligent road stud array according to claim 3, characterized in that: The eigenvectors in the vehicle position signal are automatically extracted according to the denoising autoencoder model, specifically: Vehicle position signal sequence preprocessing: The original vehicle position signal sequence X is converted into a new sequence X with a length of 100 by using the linear interpolation method. 100 . Feature extraction: A one-dimensional convolutional neural network is used to extract local features of the standardized waveform sequence to obtain vector Y; Y=Conv1d(X 100 ) Then, the vector Y is fused and reduced in dimension through the fully connected layer to obtain the feature vector F with a length of the set value S. S , F S = FC(Y).

5. The vehicle tracking method based on intelligent road stud array according to claim 1, characterized in that: The dynamic time warping algorithm is used to calculate the similarity of the two feature vectors, and the similarity is used to quantify the association weights of multiple vehicle position signal sequences collected by the intelligent road stud units in adjacent cross sections, specifically: Initialization: Create a two-dimensional dynamic programming matrix M to store local distances. The size of M is equal to the length of the two feature sequences. Create a matrix C of the same shape to store the minimum cumulative distance. Calculate the local distance: For two feature sequences in the feature vector F a and F b For each pair of data points in , the local distance is calculated using the Euclidean distance: M(i,j)=||F a (i)-F b (j)||1≤i,j≤S Among them, F a (i) is the feature sequence F a The i-th data point, F b (j) is the feature sequence F b The jth data point; M(i,j) corresponds to the feature sequence F a The i-th data point and the feature sequence F b The local distance of the jth data point; Dynamic programming filling: Starting from the upper left corner of the matrix M, calculate the minimum cumulative distance C for each position i,j , which is the sum of the current local distance and the minimum of the left, top, and top-left adjacent elements: C i,j =min{C i-1,j-1 ,C i-1,j ,C i,j-1 }+M(i,j) Returns the minimum cumulative distance: The element in the lower right corner of the matrix represents the minimum cumulative distance, denoted as DTW(F a ,F b ), represents the optimal registration distance between two sequences, where the smaller the optimal registration distance, the higher the similarity.

6. The vehicle tracking method based on intelligent road stud array according to claim 1, characterized in that: According to the Kuhn-Munkres algorithm, the vehicle position signal sequences of adjacent cross sections are optimally associated to achieve the matching of two position signal sequences of the same vehicle in adjacent cross sections to form the vehicle trajectory. The adjacent cross sections are specifically determined as follows: Time and space constraints: The time difference between vehicles passing through adjacent sections should satisfy: Among them, t i and t j represents the time when the vehicle passes through two consecutive cross sections, Δt low and Δt up are the lower and upper limits of the time difference; Spatial constraint: The number of lanes that a vehicle crosses between two adjacent sections should satisfy: Among them, l i and l j Represents the identification number of the lane where the two vehicle position signal sequences are located, Δl up Represents the maximum number of lane changes allowed.

7. A vehicle tracking method based on an intelligent road stud array according to claim 6, characterized in that: According to the Kuhn-Munkres algorithm, the vehicle position signal sequences of adjacent cross sections are optimally associated to achieve the matching of two position signal sequences of the same vehicle in adjacent cross sections to form the vehicle trajectory, which is specifically: Weight w i,j Reflecting the vehicle position signal sequence o i and the vehicle position signal sequence d j The association between them, where the larger the weight, the higher the probability of association between them; The inverse of the minimum cumulative distance is used as the weight, so o i and d j The weight w between i,j : Weight matrix W: Finally, the Kuhn-Munkres algorithm is used to determine the maximum weight matching set R: The optimal association between vehicle position signal sequences is achieved, and the matching of two position signal sequences of the same vehicle in adjacent cross sections is achieved.

8. A vehicle tracking system based on an intelligent road stud array, characterized in that: include: Vehicle position signal sequence acquisition unit: acquires the vehicle position signal sequence by using intelligent road stud units arranged along lane markings; Feature vector extraction unit: automatically extracts feature vectors from the vehicle position signal sequence based on the denoising autoencoder model; Similarity calculation unit: a dynamic time warping algorithm is used to calculate the similarity of two feature vectors, and the similarity is used to quantify the association weights of multiple vehicle position signal sequences collected by the intelligent road stud units in adjacent cross sections; Vehicle position signal sequence association unit: According to the Kuhn-Munkres algorithm, the vehicle position signal sequences of adjacent cross sections are optimally associated to achieve the matching of two position signal sequences of the same vehicle in adjacent cross sections to form a vehicle trajectory.

9. An electronic device, characterized in that: include: A memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the method steps of any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps of any one of claims 1 to 7 are implemented.

Images