Human-vehicle association analysis method based on spatio-temporal feature fusion

Through the human-vehicle correlation analysis method of fusion of space-time features, the graph attention network and time convolution long and short-term memory network are used to construct the space-time relationship diagram of vehicles and people, and the accuracy and real-time nature of the human-vehicle correlation analysis in the existing technology is solved. The problem has achieved more efficient human-vehicle correlation analysis.

CN119066503AActive Publication Date: 2024-12-03HANGZHOU DIANZI UNIV
View PDF 8 Cites 0 Cited by

Patent Information

Application Number
CN202411146526.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-20
Publication Date
2024-12-03
Estimated Expiration
2044-08-20

AI Technical Summary

Technical Problem

The existing human-vehicle association methods have poor accuracy and robustness when dealing with complex spatiotemporal relationships, and video analysis methods consume a lot of computing resources, and lack real-time and environmental adaptability.

Method used

The human-vehicle correlation analysis method based on spatiotemporal and spatial characteristics fusion is adopted, and the spatial and temporal characteristics are captured through the graph attention network and the temporal convolutional length and short-term memory network is captured by constructing the space-time relationship between vehicles and people, and calculating the cosine similarity for correlation analysis.

Benefits of technology

It improves the accuracy of human-vehicle correlation analysis, can capture human-vehicle relationship more comprehensively, and enhances the accuracy and real-time analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066503A_ABST
    Figure CN119066503A_ABST
Patent Text Reader

Abstract

The invention discloses a man-vehicle association analysis method based on spatio-temporal feature fusion, and the method comprises the following steps: respectively collecting the license plate number of a vehicle and the IMSI code of a mobile phone by using two different devices for all checkpoints in a target road network; the collected data are preprocessed, and license plate numbers and IMSI codes collected at the same checkpoint in the same time period form a sample; respectively constructing a graph for each sample, and taking a license plate number and an IMSI code as nodes; capturing spatial correlation characteristics of each license plate number or IMSI code in each time slice by using a graph attention network, capturing time correlation characteristics of the license plate number and the IMSI code in different time slices by using a time convolution long short-term memory network, and generating final vector representation of the license plate number and the IMSI code; and finally, calculating the similarity between all license plate numbers and IMSI code vectors, and selecting the IMSI code or license plate number with the highest similarity as an association result. According to the invention, the space-time correlation between the license plate number and the IMSI code can be effectively captured, and the accuracy of human-vehicle correlation analysis is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of spatiotemporal data processing, and in particular to a method for analyzing human-vehicle association based on spatiotemporal feature fusion. Background Art

[0003] There are two main types of current methods for associating people with vehicles. The first is the early methods that rely on predefined rules and simple statistical features to perform association analysis to build the connection between license plate numbers and mobile phone numbers. These methods usually have difficulty handling complex spatiotemporal relationships and have poor accuracy and robustness. The second is to extract vehicle images and face images from the video through video analysis, and then obtain vehicle information and person identity information based on the vehicle images and face images respectively to associate people with vehicles. This type of method requires a lot of computing resources to process video data and the process is complex, with poor real-time and environmental adaptability. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention proposes a human-vehicle association analysis method based on spatiotemporal feature fusion, which effectively integrates spatial features and temporal features, aiming to improve the accuracy of human-vehicle association.

[0005] In order to solve the above technical problems, the technical solution of the present invention is:

[0006] A method for analyzing human-vehicle association based on spatiotemporal feature fusion, comprising the following steps:

[0007] S1. For all checkpoints in the target road network, two different devices are used to collect the vehicle license plate number and the IMSI code of the mobile phone respectively, to form license plate number related data and IMSI code related data; the license plate number related data includes the checkpoint number, license plate number, and collection time, and the IMSI code related data includes the checkpoint number, IMSI code, and collection time;

[0008] S2, pre-processing the collected license plate number related data and IMSI code related data to filter out missing data, redundant data, outlier data and non-related data;

[0009] S3, dividing the pre-processed license plate numbers and IMSI codes according to the checkpoint number and collection time, and the license plate numbers and IMSI codes collected by the same checkpoint in the same time period form a sample, and all samples form a sample set;

[0010] S4, treat each sample as a document, use the word embedding model to train, and obtain the initial word embedding of each license plate number and IMSI code;

[0011] S5. Construct a graph for each sample, take each license plate number and IMSI code as a node, and use the word embedding of the license plate number and IMSI code as the initial feature vector of the corresponding node; establish an edge between every two nodes, and calculate the weight of the edge according to the time difference between the collection of the two node data;

[0012] S6. For each license plate number, obtain a graph containing the license plate number node in each time slice; for each obtained graph, use a graph attention network to capture the spatial features of the node, and obtain a new feature vector of the license plate number node in each time slice;

[0013] S7, taking the feature vector corresponding to the license plate number in each time slice as the time series input, passing it to the time convolutional long short-term memory network to capture the time characteristics of the node, and obtaining the hidden state of the license plate number at each time step;

[0014] S8. Use the average pooling method to aggregate the hidden states of each time step to obtain the final vector representation of each license plate number;

[0015] S9, repeating steps S6-S8, for each IMSI code, obtaining a graph containing the IMSI code node in each time slice, using the graph attention network to extract the spatial features of the node, and obtaining a new feature representation of the IMSI code node in each time slice; then, inputting the new feature vector of each time slice as a time series into the time convolutional long short-term memory network to extract the time features of the node, and obtaining the hidden state of the IMSI code at each time step; then, using the average pooling method to aggregate the hidden states output at each time step, and obtaining the final vector representation of each IMSI code;

[0016] S10, calculating the cosine similarity of the final vector representations of all license plate numbers and IMSI codes; for any license plate number or IMSI code, selecting the IMSI code or license plate number with the highest cosine similarity as its association result.

[0017] Preferably, the data preprocessing method process is as follows:

[0018] S211. If any field value of the checkpoint number, license plate number or IMSI code, or collection time in a piece of data is missing, directly delete the piece of data;

[0019] S212. For data collected at the same checkpoint in the same time period but with different collection time field values ​​but the same license plate number or IMSI code field values, retain the data with the earliest collection time and delete the remaining redundant data;

[0020] S213. If the license plate number or IMSI code in the data does not conform to the rules, filter out the data; similarly, if the collection time field value exceeds the set time range, filter out the data that exceeds the range;

[0021] S214. If a license plate number or IMSI code appears only once in the data collected by all checkpoints, delete all data related to the license plate number or IMSI code;

[0022] S215. The pre-processed data is used to form a new license plate number related data set and an IMSI code related data set.

[0023] Preferably, the data division method starts from 00:00:00, with each minute being a time period, and the license plate number and IMSI code collected by each checkpoint in each time period are combined into one sample data.

[0024] Preferably, the weight calculation method of the edge in the graph structure is that for any pair of nodes p and q, the weight of the edge (p, q) can be expressed as where t p and t q They represent the collection time of nodes p and q respectively, in seconds, and finally the weight matrix W of each graph is obtained.

[0025] Preferably, the specific process of the spatial feature extraction method is as follows:

[0026] S611, each node i in the graph has an initial feature vector h′ i , these feature vectors are transformed using a learnable weight matrix M:

[0027] h″ i =Mh′ i

[0028] S612: For each node j connected to node i, the weight of the fused edge is used to calculate the attention coefficient e ij :

[0029] e ij =LeakyReLU(a T [h″ i ||h″ j ||w ij ])

[0030] Among them, a T is a learnable weight vector, || represents vector concatenation, LeakyReLU is an activation function with a negative slope, and w ij is the weight of edge (i, j);

[0031] S613, use Softmax function to adjust the attention coefficient e ij Normalize to ensure that the sum of the attention coefficients of all neighbors of node i is 1;

[0032] S614: Use the normalized attention coefficient to perform weighted summation on the features of neighboring nodes to obtain a new feature vector x of node i i ;

[0033] S615. Repeat S611-S614 to obtain a new vector representation of the license plate number in each time slice.

[0034] Preferably, the temporal convolutional long short-term memory neural network is composed of a stack of temporal convolutional layers and long short-term memory layers; the input of the temporal convolutional layer is a time series composed of new feature vectors in each time slice of the license plate number, and an expanded convolutional layer is used in the layer to perform a convolution operation on the time series data to extract local features in the time series, and then output the new feature vector of the license plate number at each time step and pass it to the next layer; the input of the long short-term memory layer is the output of the temporal convolutional layer, and for each time step, the input vector will be calculated through the input gate, forgetting gate, candidate memory cell and output gate to obtain the hidden state of the time step, and finally the hidden state of each time step is used as the output of the temporal convolutional long short-term memory network.

[0035] The present invention has the following characteristics and beneficial effects:

[0036] By adopting the above technical solution, a graph is constructed according to the spatial and temporal relationships between vehicles and people, so that when performing human-vehicle association analysis, not only the corresponding relationship between people and vehicles in the spatial dimension is considered, but also the corresponding relationship between people and vehicles in the temporal dimension is considered, thereby obtaining a more accurate human-vehicle association result. The present invention is based on a graph attention network and uses an attention mechanism to capture the spatial association characteristics between people and vehicles that appear in the same place; at the same time, the present invention introduces a temporal convolutional long short-term memory network to capture the temporal association characteristics between people and vehicles in different time slices. By integrating the spatial and temporal characteristics of people and vehicles, the present invention can capture the relationship between people and vehicles more comprehensively and effectively improve the accuracy of human-vehicle association analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] 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 labor.

[0038] Figure 1This is a flow chart of the human-vehicle association analysis method based on spatiotemporal feature fusion proposed by the present invention;

[0039] Figure 2 Schematic diagram of data preprocessing process;

[0040] Figure 3 It is a schematic diagram of the spatial feature extraction process;

[0041] Figure 4 This is a structural diagram of the human-vehicle association analysis method based on the fusion of spatiotemporal features. DETAILED DESCRIPTION

[0042] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0043] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings.

[0044] On the contrary, the present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention as defined by the claims. Further, in order to make the public have a better understanding of the present invention, some specific details are described in detail in the detailed description of the present invention below. Those skilled in the art can fully understand the present invention without the description of these details.

[0045] In a preferred embodiment of the present invention, a method for analyzing human-vehicle association based on spatiotemporal feature fusion is provided. Figure 1 and Figure 4 The specific steps are shown in S1 to S10. In this method, in addition to using the graph attention network to capture the spatial features of the license plate number and IMSI code at different checkpoints, the time convolutional long short-term memory network is also used to mine the temporal features of the license plate number and IMSI code at different time slices, and the accurate association between people and cars is achieved by combining the spatial and temporal features of the data. Figure 1 As shown, the specific implementation of the above steps S1 to S10 is described in detail below.

[0046] S1. For all checkpoints in the target road network, two different devices are used to collect the vehicle license plate number and the mobile phone IMSI code respectively to form license plate number related data and IMSI code related data; the license plate number related data includes the checkpoint number, license plate number, and collection time, and the IMSI code related data includes the checkpoint number, IMSI code, and collection time.

[0047] It should be noted that the above-mentioned license plate number is used to refer to the vehicle, and the IMSI code of the mobile phone is used to refer to the person, and the association between the license plate number and the IMSI code is realized by associating the person and the vehicle; the data collection time lasts for one week, that is, starting from the specified start time, data collection is carried out around the clock for 7 consecutive days; the license plate number data of the vehicle at the checkpoint and the IMSI code data of the mobile phone can be obtained by the corresponding equipment or sensors installed on the road in actual application scenarios, which belongs to the existing technology and will not be described in detail.

[0048] S2. Preprocess the collected license plate number related data and IMSI code related data to filter out missing data, redundant data, outlier data and non-related data.

[0049] Specifically, Figure 2 As shown, the process is as follows:

[0050] S211. If any field value of the checkpoint number, license plate number or IMSI code, or collection time in a certain piece of data is missing, the data is directly deleted.

[0051] S212. For data collected at the same checkpoint in the same time period but with different collection time field values ​​but the same license plate number or IMSI code field values, retain the data with the earliest collection time and delete the remaining redundant data.

[0052] S213. If the license plate number or IMSI code in the data does not comply with the rules, filter out the data; similarly, if the collection time field value exceeds the set time range, filter out the data that exceeds the range.

[0053] It should be noted that the ordinary license plate number is a combination of 6 to 8 Chinese characters, uppercase letters and numbers. The first character is the Chinese abbreviation of the province, the second character is a letter, and the rest are numbers or letters; the IMSI code is 15 digits, the first 3 digits are the legal mobile country code (MCC), the next 2 to 3 digits are the mobile network code (MNC), and the rest are the user identification code.

[0054] S214. If a certain license plate number or IMSI code appears only once in the data collected by all checkpoints, all data related to the license plate number or IMSI code are deleted.

[0055] S215, the pre-processed data is used to form a new license plate number related data set and an IMSI code related data set, which are respectively represented as C′=(c1, c2, ..., c n′ ) and P′=(p1,p2,...,p m′ ), n′ and m′ represent the number of pre-processed license plate number and IMSI code related data.

[0056] S3, divide the pre-processed license plate number and IMSI code according to the checkpoint number and collection time, and form a sample with the license plate number and IMSI code collected by the same checkpoint in the same time period. Starting from 00:00:00, each 1 minute is a time period. The license plate number and IMSI code collected by each checkpoint in each time period are combined into a sample data, forming a total of T×K samples, where T is the total number of time slices, K is the total number of checkpoints, and the set formed by all samples can be expressed as B=(b1, b2, ..., b T ), where b i =(b i,1 , b i,2 , ..., b i,k ) represents the sample set of the i-th time slice, b i,j Represents a sample.

[0057] It should be noted that each sample only contains the license plate number and IMSI code, and does not include the checkpoint number and collection time.

[0058] S4. Treat each sample as a document and use the word embedding model (Word2Vec) for training to obtain the initial word embedding of each license plate number and IMSI code.

[0059] S5. Construct a graph for each sample, with the license plate number and IMSI code as nodes, and the word embedding of the license plate number and IMSI code as the initial feature vector of the corresponding node; establish an edge between every two nodes, and calculate the weight of the edge based on the time difference between the collection of the two node data.

[0060] It should be noted that the construction of the graph belongs to the existing technology of Graph Neural Networks (GNNs), and can be implemented by referring to the existing technology. In the embodiment of the present invention, the weight calculation method of the edge in the graph structure is that for any pair of nodes p and q, the weight of the edge (p, q) can be expressed as where t p and t q They represent the collection time of nodes p and q respectively, in seconds, and finally the weight matrix W of each graph is obtained.

[0061] S6. For each license plate number, obtain the graph containing the license plate number node in each time slice; for each obtained graph, use the graph attention network to capture the spatial features of the node to obtain a new feature vector of the license plate number node in each time slice.

[0062] Specifically, Figure 3 As shown, the process is as follows:

[0063] S611, each node i in the graph has an initial feature vector h′i , these feature vectors are transformed using a learnable weight matrix M:

[0064] h″ i =Mh′ i °

[0065] S612: For each node j connected to node i, the weight of the fused edge is used to calculate the attention coefficient e ij :

[0066] e ij =LeakyReLU(a T [h″ i ||h″ j ||w ij ])

[0067] Among them, a T is a learnable weight vector, || represents vector concatenation, LeakyReLU is an activation function with a negative slope, and w ij is the weight of edge (i, j).

[0068] S613, use Softmax function to adjust the attention coefficient e ij Normalization is performed to ensure that the sum of the attention coefficients among all neighbors of node i is 1:

[0069]

[0070] Among them, N(i) represents the set of neighbor nodes of node i.

[0071] S614: Use the normalized attention coefficient to perform weighted summation on the features of neighboring nodes to obtain a new feature vector x of node i i :

[0072]

[0073] Where σ is the activation function ReLU.

[0074] S615, calculate the vector representation of the license plate number node in each graph, so as to obtain the new vector representation x1, x2, ..., x of the license plate number in each time slice. T , where T is the total number of time slices.

[0075] It should be noted that the Graph Attention Network (GAT) also belongs to the prior art. Its function is to use the attention mechanism to calculate the relevance weight of each node with its neighboring nodes, and dynamically calculate the attention weight during the training process, and then generate a new feature representation of each node by aggregating the features of the neighboring nodes and weighting them according to their importance, which can effectively capture the complex relationship between the license plate number and the IMSI code. In an embodiment of the present invention, the weight of the edge is fused into the calculation of the GAT attention coefficient, so that the model can better capture the important information in the graph.

[0076] S7, taking the feature vector corresponding to the license plate number in each time slice as the time series input, passing it to the time convolutional long short-term memory network to capture the time characteristics of the node, and obtaining the hidden state of the license plate number at each time step;

[0077] It should be noted that the temporal convolutional network-long short-term memory network (TCN-LSTM) is composed of a stack of temporal convolutional layers (TCN) and long short-term memory layers (LSTM); the input of the temporal convolutional layer is a time series consisting of feature vectors corresponding to the license plate number in each time slice. A one-dimensional convolutional layer is used in the layer to perform convolution operations on the time series data, extract local features in the time series, and then output the new feature vector of the license plate number at each time step and pass it to the next layer; the input of the long short-term memory layer is the output of the temporal convolutional layer. For each time step, the input vector will be calculated through the input gate, forget gate, candidate memory cell and output gate to obtain the hidden state of the time step, and finally the hidden state of each time step is used as the output of the temporal convolutional long short-term memory network. The specific process is as follows:

[0078] S711, the characteristic vector x of the license plate number at each time step t , combined in chronological order into a time series data X = {x1, x2, ..., x T}, where T is the total number of time slices.

[0079] S712: Input the time series data X into the TCN layer, and after causal convolution and dilated convolution, obtain the new feature vector d at each time step. t :

[0080]

[0081] Where W and b are the convolution kernel and bias term respectively, K is the convolution kernel size, g is the dilation factor, and f is the ReLU activation function.

[0082] S713: The output sequence D of the TCN layer is T} As the input of the LSTM layer, construct input gate, forget gate and output gate units to process the input:

[0083] i t =σ(W i ·[h t-1 , d t ]+b i )

[0084] f t =σ(W f ·[h t-1 , d t ]+b f )

[0085] o t =σ(W o ·[h t-1 , d t ]+b o )

[0086]

[0087]

[0088] h t =o t *tanh(C t )

[0089] where h t is the hidden state of the current time step, i t , f t , o t are the activation values ​​of the input gate, forget gate, and output gate, respectively. are candidate memory cells, C t is the memory cell state, W i , W f , W o , W C They are the weight matrices of the input gate, forget gate, output gate and candidate memory cells respectively.

[0090] S714. The model finally outputs the hidden state H corresponding to each time step = {h1, h2, ..., h T}, as the feature vector of the license plate number at each time step.

[0091] S8. Use the average pooling method to aggregate the hidden states of each time step to obtain the final vector representation v of each license plate number. c :

[0092]

[0093] where h t [c] is the hidden state vector at time step t; finally, we get the vector representation of all license plate numbers

[0094] S9. Repeat steps S6-S8. For each IMSI code, obtain the graph containing the IMSI code node in each time slice, use the graph attention network to extract the spatial features of the node, and obtain the new feature representation of the IMSI code node in each time slice; then, input the new feature vector of each time slice as a time series into the time convolutional long short-term memory network to extract the time features of the node, and obtain the feature vector of the IMSI code at each time step; then use the average pooling method to aggregate the feature vectors output from each time step to obtain the final vector representation of each IMSI code.

[0095] S10, calculating the cosine similarity of the final vector representations of all license plate numbers and IMSI codes; for any license plate number or IMSI code, selecting the IMSI code or license plate number with the highest cosine similarity as its association result.

[0096] It should be noted that the cosine similarity calculation method, for any license plate number c x and IMSI code p y The corresponding vector representation and calculate and The cosine similarity between

[0097]

[0098] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions and variations of these embodiments including components are made without departing from the principles and spirit of the present invention, and still fall within the scope of protection of the present invention.

Claims

1. A human-vehicle association analysis method based on spatiotemporal feature fusion, characterized in that: The steps include: Step 1: For all checkpoints in the target road network, collect the vehicle license plate number related data and the mobile phone IMSI code related data respectively, and pre-process them; Step 2: Divide the license plate number and IMSI code in the pre-processed relevant data according to the card port number and collection time to obtain a sample set; Step 3: Treat each sample in the sample set as a document and use the word embedding model for training to obtain the initial word embedding of each license plate number and IMSI code; Step 4: For each sample, a graph is constructed based on the license plate number, IMSI code, and the initial word embedding of the license plate number and IMSI code, and a new feature vector of the license plate number in each time slice of the node in the graph is obtained using the graph attention network for each license plate number; Step 5: Take the feature vector of each time slice of the license plate number as the time series input and pass it to the time convolution long short-term memory network to capture the time characteristics of the node and obtain the hidden state of the license plate number at each time step; Step 6: Use the average pooling method to aggregate the hidden states of each time step to obtain the final vector representation of each license plate number; Step 7: Obtain the final vector representation of each IMSI code according to steps 4-6 for the IMSI code; Step 8: Calculate the cosine similarity of the final vector representations of all license plate numbers and IMSI codes; for any license plate number or IMSI code, select the IMSI code or license plate number with the highest cosine similarity as its association result.

2. According to the method of human-vehicle association analysis based on spatiotemporal feature fusion according to claim 1, it is characterized in that: The license plate number related data includes the checkpoint number, license plate number, and collection time, and the IMSI code related data includes the checkpoint number, IMSI code, and collection time.

3. The method for analyzing human-vehicle association based on spatiotemporal feature fusion according to claim 1 is characterized in that: The preprocessing method includes filtering out missing data, redundant data, outlier data and non-related data.

4. The method for analyzing human-vehicle association based on spatiotemporal feature fusion according to claim 1 is characterized in that: In step 2, the sample set is divided into: the license plate number and IMSI code collected at the same checkpoint in the same time period are grouped into one sample, and all samples form a sample set.

5. The method for analyzing human-vehicle association based on spatiotemporal feature fusion according to claim 4 is characterized in that: In step 2, when dividing the samples, the calculation starts from 00:00:00, and each minute is a time period.

6. The method for analyzing human-vehicle association based on spatiotemporal feature fusion according to claim 1 is characterized in that: The graph construction method in step 4 is as follows: for each sample, each license plate number and IMSI code in the sample is used as a node, and the word embedding of the license plate number and IMSI code is used as the initial feature vector of the corresponding node; an edge is established between every two nodes, and the weight of the edge is calculated according to the time difference between the collection of the two node data.

7. The method for analyzing human-vehicle association based on spatiotemporal feature fusion according to claim 6 is characterized in that: The weight calculation method is: For any pair of nodes p and q, the weight of the edge (p, q) can be expressed as where t p and t q They represent the collection time of nodes p and q respectively, in seconds, and finally the weight matrix W of each graph is obtained.

8. The method for analyzing human-vehicle association based on spatiotemporal feature fusion according to claim 7 is characterized in that: The step 4 also includes obtaining, for each license plate number, a graph containing the license plate number node in each time slice; for each obtained graph, using a graph attention network to capture the spatial features of the node, and obtaining a new feature vector of the license plate number node in each time slice.

9. The method for analyzing human-vehicle association based on spatiotemporal feature fusion according to claim 7 is characterized in that: In step 4, the method for obtaining the new feature vector of the license plate number node in each time slice is as follows: Each node i in the graph has an initial feature vector h′ i , use a learnable weight matrix M to transform these feature vectors; h″ i =Mh′ i For each node j connected to node i, the weight of the fused edge is used to calculate the attention coefficient e ij : and ij =LeakyReLU(a T [h″ i ||h″ j ||in ij ]) Among them, a T is a learnable weight vector, || represents vector concatenation, LeakyReLU is an activation function with a negative slope, and w ij is the weight of edge (i, j); Use the Softmax function to adjust the attention coefficient e ij Normalize to ensure that the sum of the attention coefficients of all neighbors of node i is 1; Use the normalized attention coefficient to weight the features of neighboring nodes and get the new feature vector x of node i i ; Repeat the above steps to obtain a new feature vector of the license plate number in each time slice.

10. The method for analyzing human-vehicle association based on spatiotemporal feature fusion according to claim 9 is characterized in that: In step 7, when obtaining new feature representations of the IMSI code nodes in each time slice, for each IMSI code, a graph containing the IMSI code node in each time slice is obtained.

11. The method for analyzing human-vehicle association based on spatiotemporal feature fusion according to claim 1, characterized in that: In step 5, the temporal convolutional long short-term memory network is composed of a stack of temporal convolutional layers and long short-term memory layers; the input of the temporal convolutional layer is a time series composed of new feature vectors of the license plate number in each time slice, and an expanded convolutional layer is used in the layer to perform a convolution operation on the time series data to extract local features in the time series, and then output the vector representation of the license plate number at each time step and pass it to the next layer; the input of the long short-term memory layer is the output of the temporal convolutional layer, and for each time step, the input vector will be calculated through the input gate, the forgetting gate, the candidate memory cell and the output gate to obtain the hidden state of the time step, and finally the hidden state of each time step is used as the output of the temporal convolutional long short-term memory network.

Citation Information

Patent Citations

  • Geohash matching-based vehicle code fitting method and system

    CN110888884A

  • Solar photovoltaic power generation prediction method based on TCN-LSTM

    CN110909926A

  • Vehicle trajectory prediction method based on LSTM and space-time attention mechanism

    CN114372116A

  • Human-vehicle association analysis method based on word vectors

    CN115690766A

  • Emotion analysis method based on BiLSTM-GAT model

    CN117609494A