Abnormal trajectory detection method and system based on recurrent neural network and differential autoencoder

Abstract

The present invention discloses an abnormal trajectory detection method and system based on a recurrent neural network and a differential autoencoder, comprising: segmenting the collected original trajectory data to obtain segmented trajectory data; dividing the target map into a number of grids, discretizing the segmented trajectory data to obtain discretized trajectory data; dividing the discretized trajectory data to obtain a training set and a test set; constructing an abnormal trajectory detection model based on a recurrent neural network and a differential autoencoder; inputting the training set into the abnormal trajectory detection model for training to obtain an optimized abnormal trajectory detection model; inputting the test set into the optimized abnormal trajectory detection model to obtain an abnormal score of the trajectory; judging whether the trajectory is abnormal based on the abnormal score. The present invention can encode the trajectory information of the vehicle by generating a model, explore the manifestation of trajectory anomalies in the latent space, and discriminate the anomalies to improve the accuracy of detection.

Description

Technical Field

[0001] The present invention belongs to the technical field of trajectory detection, and relates to an abnormal trajectory detection method and system based on a recurrent neural network and a differential autoencoder. Background Art

[0002] The rapid development of global satellite positioning navigation systems (GPS) and handheld mobile terminals has promoted the progress of research based on the location information of mobile objects. Due to the increase in portability and the decrease in cost, a large amount of trajectory data of mobile objects has been collected and stored. There is a strong temporal and spatial correlation between these data, and these data with spatiotemporal characteristics characterize the movement characteristics of the object. However, the implicit local patterns are often complex, and it is difficult to mine effective information from the data. Therefore, it is necessary to establish an effective data management mechanism, study data mining technology in big data, and extract valuable spatiotemporal patterns and other effective information from the regular data of a large number of mobile objects.

[0003] Existing research on abnormal trajectory detection can be divided into two categories: metric-based methods and learning-based methods. The first category is usually based on manual features. Since trajectories between different locations are usually very different, these definitions may only be valid for a few cases, but not for many other cases. Secondly, trajectory comparison is usually time-consuming and cannot support effective online detection. Many deep learning methods have been applied to the anomaly detection problem, however, they lack the means to mine useful information about trajectory anomalies in the latent embedding space, that is, they can only embed the trajectory into a low-dimensional space without exploring which part of the latent space represents the normal and which part represents the anomaly. Therefore, the use of probabilistic models provides a new idea for modeling abnormal trajectories. Summary of the invention

[0004] The purpose of the present invention is to solve the problems in the prior art and provide an abnormal trajectory detection method and system based on recurrent neural network and differential autoencoder, which can effectively improve the accuracy of abnormal trajectory detection.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] The abnormal trajectory detection method based on recurrent neural network and differential autoencoder includes:

[0007] Segment the collected original trajectory data to obtain segmented trajectory data;

[0008] Divide the target map into several grids, discretize the segmented trajectory data, and obtain discretized trajectory data;

[0009] Divide the discretized trajectory data to obtain training sets and test sets;

[0010] Based on recurrent neural network and differential autoencoder, an abnormal trajectory detection model is constructed;

[0011] Input the training set into the abnormal trajectory detection model for training to obtain the optimized abnormal trajectory detection model;

[0012] Input the test set into the optimized abnormal trajectory detection model to obtain the abnormal score of the trajectory;

[0013] Based on the anomaly score, determine whether the trajectory is abnormal.

[0014] A further improvement of the present invention is:

[0015] The collected original trajectory data is segmented to obtain the segmented trajectory data, specifically:

[0016] The vehicle's trajectory point location information and timestamp are obtained from the driving trajectory data; based on the vehicle's trajectory point location information and timestamp, the sequence P = {p1, p2, ..., p n} represents a trajectory in the trajectory data; where p i =(t i ,l i )∈P represents the timestamp and position information of the i-th trajectory point in the sequence P, and n is the length of the trajectory;

[0017] Based on the trajectory segmentation rule, the original trajectory data P is segmented using the sequence Q = {q1, q2, …, q m} represents a trajectory in the trajectory data; where q i =(t i ,l i )∈Q represents the timestamp and position information of the i-th trajectory point in the sequence Q, and m is the length of the trajectory.

[0018] The trajectory segmentation rules are as follows:

[0019] When the spatial deviation of two adjacent trajectory points is greater than the set first threshold, Lagrange interpolation is performed through the surrounding trajectory points to form a complete trajectory data; or when the trajectory point does not move within the set time or the displacement deviation of the trajectory point does not reach the set second threshold, the trajectory is segmented to form a complete trajectory data;

[0020] The trajectory is segmented as follows:

[0021] The trajectory points that have not moved within the set time are eliminated, and the non-moving displacement points are segmented to obtain the segmented trajectory data.

[0022] The target map is divided into several grids, and the segmented trajectory data is discretized to obtain the discretized trajectory data, specifically:

[0023] The target map is divided into several grids. The trajectory points falling into the same grid have the same number to discretize the trajectory data points. The sequence R = {r1, r2, ..., r m} represents a trajectory in the trajectory data; where r i =(t i ,l i )∈R represents the timestamp and position information of the ith trajectory point in the sequence R, and m is the length of the trajectory.

[0024] Based on recurrent neural network and differential autoencoder, an abnormal trajectory detection model is constructed, specifically:

[0025] Construct a trajectory autoencoder model; including encoder E1, probability distribution P1 and generator G1; encoder E1 is a recurrent neural network, which takes the trajectory data R1 in the training set as input and outputs the potential representation r of the trajectory R ; The probability distribution P1 of the path is a probability distribution model based on Gaussian distribution, with the potential representation r of the trajectory R As input, output μ={μ1,μ2,…,μ n} represents the path type in the latent space; where n is the number of types; the generator G1 is a recurrent neural network that transforms the potential representation of the trajectory r R As input, output is the reconstructed trajectory

[0026] Construct a discriminator D; the discriminator D is a k-layer fully connected neural network that combines the trajectory data R1 in the training set with the reconstructed trajectory As input, the output score matrix

[0027] The training set is input into the abnormal trajectory detection model for training to obtain the optimized abnormal trajectory detection model, specifically:

[0028] Step 1: Use trajectory information to vary the lower bound As the loss function, the trajectory data R1 in the training set and the trajectory potential representation μ are used to train the neural network parameters in the encoder E1 and the generator G1 of the user variational autoencoder using the stochastic gradient descent method; the trajectory potential representation μ;

[0029]

[0030] in, θ1 represents the neural network parameters in the encoder E1 and generator G1 to be trained; Representation trajectory The probability of belonging to the normal path;

[0031] Step 2: Repeat step 1 to train the abnormal trajectory detection model until the loss function is lower than the set threshold.

[0032] Input the test set into the optimized abnormal trajectory detection model to obtain the abnormal score of the trajectory, specifically:

[0033] The trajectory data R2 in the test set is input into the encoder part of the optimized abnormal trajectory detection model, and the output is the potential representation r of the trajectory R ;

[0034] Based on the latent representation R Calculate the path μ in the latent space k ∈μ, and calculate the anomaly score s(R) of the trajectory by the probability.

[0035] The calculation method of the anomaly score s(R) is:

[0036]

[0037] Among them, μ c Represents the path latent representation in the latent vector space.

[0038] The abnormal trajectory detection system based on recurrent neural network and differential autoencoder includes:

[0039] A segmentation module, wherein the segmentation module is used to segment the collected original trajectory data to obtain segmented trajectory data;

[0040] A first division module, wherein the first division module is used to divide the target map into a plurality of grids, discretize the segmented trajectory data, and obtain discretized trajectory data;

[0041] A second partitioning module, the second partitioning module is used to partition the discretized trajectory data to obtain a training set and a test set;

[0042] A building module, wherein the building module builds an abnormal trajectory detection model based on a recurrent neural network and a differential autoencoder;

[0043] A training module, wherein the training module is used to input the training set into the abnormal trajectory detection model for training to obtain an optimized abnormal trajectory detection model;

[0044] A testing module, wherein the testing module is used to input a test set into an optimized abnormal trajectory detection model to obtain an abnormal score of the trajectory;

[0045] A judgment module is used to judge whether the trajectory is abnormal based on the abnormality score.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention extracts the trajectory data of the vehicle, performs segmentation and discretization processing, and constructs and trains an abnormal trajectory detection model to determine whether the trajectory is abnormal. The present invention can encode the trajectory information of the vehicle through a generative model, explore the manifestation of trajectory anomalies in the latent space, and distinguish the anomalies based on a probability model; the present invention makes fuller use of known information through training, so that the obtained trajectory potential representation contains more complete information, thereby achieving a higher accuracy rate of anomaly detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 It is a flowchart of an abnormal trajectory detection method based on recurrent neural network and differential autoencoder;

[0050] Figure 2 This is the architecture diagram of the abnormal trajectory detection method based on recurrent neural network and differential autoencoder;

[0051] Figure 3 This is a flow chart for path probability inference;

[0052] Figure 4 This is the structural diagram of the abnormal trajectory detection system based on recurrent neural network and differential autoencoder. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0054] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. 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.

[0055] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0056] In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. indicate an orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0057] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0058] In the description of the embodiments of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0059] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0060] See also Figure 1 and Figure 2 The present invention discloses an abnormal trajectory detection method based on a recurrent neural network and a differential autoencoder, comprising:

[0061] S101, segmenting the collected original trajectory data to obtain segmented trajectory data.

[0062] The vehicle's trajectory point location information and timestamp are obtained from the driving trajectory data; based on the vehicle's trajectory point location information and timestamp, the sequence P = {p1, p2, ..., p n} represents a trajectory in the trajectory data; where p i =(t i ,l i )∈P represents the timestamp and position information of the i-th trajectory point in the sequence P, and n is the length of the trajectory;

[0063] Based on the trajectory segmentation rule, the original trajectory data P is segmented using the sequence Q = {q1, q2, …, q m} represents a trajectory in the trajectory data; where q i =(t i ,l i )∈Q represents the timestamp and position information of the i-th trajectory point in the sequence Q, and m is the length of the trajectory.

[0064] In this embodiment, the trajectory dataset is a T-Drive trajectory dataset. A taxi driving trajectory dataset is derived from:

[0065] https: / / www.microsoft.com / en-us / research / publication / t-drive-trajectory-data-sample, the dataset includes the one-week trajectories of 10,357 taxis. The total number of points in the dataset is about 15 million, and the total distance of the trajectories reaches 9 million kilometers. The basic information such as the vehicle's trajectory points and timestamps are obtained in the driving trajectory data.

[0066] Using the sequence P = {p1, p2, ..., p n}={(0,116.51,39.93),(600,116.51,39.93),(600,116.47,39.91),…,(1456,116.47,39.90)} represents a trajectory in the trajectory data, where p i =(t i ,l i )∈P represents the timestamp and position information of the i-th trajectory point in the sequence P, and n is the length of the trajectory.

[0067] The original trajectory data P is segmented according to the following segmentation rules: when the spatial deviation of two adjacent trajectory points is greater than the set first threshold, Lagrange interpolation is performed through the surrounding trajectory points to form a complete trajectory data; or when the trajectory point does not move within the set time or the displacement deviation of the trajectory point does not reach the set second threshold, the trajectory is segmented to form a complete trajectory data.

[0068] The trajectory is segmented as follows:

[0069] The trajectory points that have not moved within the set time are eliminated, and the non-moving displacement points are segmented to obtain the segmented trajectory data.

[0070] In the above sequence, the positions of p1 and p2 have not changed, indicating that the vehicle has not moved within 600 seconds. Therefore, the trajectory is segmented, the trajectory point of p1 is removed, and the sequence Q = {q1, q2, …, q m}={(600,116.51,39.93),(600,116.47,39.91),…,(1456,116.47,39.90)} represents a trajectory in the trajectory data, where q i =(t i ,l i )∈Q represents the timestamp and position information of the i-th trajectory point in the sequence Q, and m is the length of the trajectory.

[0071] S102, dividing the target map into a plurality of grids, performing discretization processing on the segmented trajectory data, and obtaining discretized trajectory data;

[0072] The target map is divided into several grids. The trajectory points falling into the same grid have the same number to discretize the trajectory data points. The sequence R = {r1, r2, …, r m} represents a trajectory in the trajectory data; where r i =(t i ,l i )∈R represents the timestamp and position information of the ith trajectory point in the sequence R, and m is the length of the trajectory.

[0073] The entire map is divided into 300m*300m grids. The trajectory points falling into the same grid have the same number to achieve discretization of trajectory data points. The 200 most passed grid points are selected as the starting point or destination to further divide the trajectory. Use the sequence R = {r1, r2, …, r m}={(600,184,35),(600,169,28),…,(1456,169,24)} represents a trajectory in the trajectory data, where r i =(t i ,l i )∈R represents the timestamp and position information of the ith trajectory point in the sequence R, and m is the length of the trajectory.

[0074] S103, dividing the discretized trajectory data to obtain a training set and a test set.

[0075] S104, constructing an abnormal trajectory detection model based on a recurrent neural network and a differential autoencoder.

[0076] Construct a trajectory autoencoder model; including encoder E1, probability distribution P1 and generator G1; encoder E1 is a recurrent neural network, which takes the trajectory data R1 in the training set as input and outputs the potential representation r of the trajectory R ; The probability distribution P1 of the path is a probability distribution model based on Gaussian distribution, with the potential representation r of the trajectory R As input, output μ={μ1,μ2,…,μ n} represents the path type in the latent space; where n is the number of types; the generator G1 is a recurrent neural network that transforms the potential representation of the trajectory r R As input, output is the reconstructed trajectory

[0077] Construct a discriminator D; the discriminator D is a k-layer fully connected neural network that combines the trajectory data R1 in the training set with the reconstructed trajectory As input, the output score matrix

[0078] S105: Input the training set into the abnormal trajectory detection model for training to obtain the optimized abnormal trajectory detection model.

[0079] Step 1: Use trajectory information to vary the lower bound As the loss function, the trajectory data R1 in the training set and the trajectory potential representation μ are used to train the neural network parameters in the encoder E1 and the generator G1 of the user variational autoencoder using the stochastic gradient descent method; the trajectory potential representation μ;

[0080]

[0081] in, θ1 represents the neural network parameters in the encoder E1 and generator G1 to be trained; Representation trajectory The probability of belonging to the normal path;

[0082] Step 2: Repeat step 1 to train the abnormal trajectory detection model until the loss function is lower than the set threshold.

[0083] S106, input the test set into the optimized abnormal trajectory detection model to obtain the abnormal score of the trajectory.

[0084] See also Figure 3 , use the trained model to infer the path probability, as follows:

[0085] The trajectory data R2 in the test set is input into the encoder part of the optimized abnormal trajectory detection model, and the output is the potential representation r of the trajectory R ;

[0086] Based on the latent representation R Calculate the path μ in the latent space k ∈μ, and calculate the anomaly score s(R) of the trajectory by the probability.

[0087] S107, judging whether the trajectory is abnormal based on the abnormality score.

[0088] Determine whether the trajectory is abnormal, specifically: compare s(R) with a specified value to determine whether the trajectory is abnormal; the specified value is a manually set value.

[0089] The calculation method of the anomaly score s(R) is:

[0090]

[0091] Among them, μ c Represents the path latent representation in the latent vector space.

[0092] See also Figure 4 The present invention discloses an abnormal trajectory detection system based on a recurrent neural network and a differential autoencoder, comprising:

[0093] A segmentation module, wherein the segmentation module is used to segment the collected original trajectory data to obtain segmented trajectory data;

[0094] A first division module, wherein the first division module is used to divide the target map into a plurality of grids, discretize the segmented trajectory data, and obtain discretized trajectory data;

[0095] A second partitioning module, the second partitioning module is used to partition the discretized trajectory data to obtain a training set and a test set;

[0096] A building module, wherein the building module builds an abnormal trajectory detection model based on a recurrent neural network and a differential autoencoder;

[0097] A training module, wherein the training module is used to input the training set into the abnormal trajectory detection model for training to obtain an optimized abnormal trajectory detection model;

[0098] A testing module, wherein the testing module is used to input a test set into an optimized abnormal trajectory detection model to obtain an abnormal score of the trajectory;

[0099] A judgment module is used to judge whether the trajectory is abnormal based on the abnormality score.

[0100] The experimental results of this embodiment are shown in Table 1:

[0101] ACC F1 SD-Pair1 0.9880 0.9796 SD-Pair2 0.9722 0.9167 SD-Pair3 1.0000 1.0000 SD-Pair4 0.9924 0.9767 SD-Pair5 0.9314 0.8955

[0102] The accuracy (ACC) of the test set is stable above 0.93, and the F1 value is stable above 0.89.

[0103] The experimental results show that the abnormal trajectory detection method based on recurrent neural network and differential autoencoder proposed in the present invention can effectively mine the useful features in the time information and spatial information of the driving trajectory, and then accurately model the abnormal pattern of the trajectory. The prediction result has high accuracy and small error, and has high practical value.

[0104] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An abnormal trajectory detection method based on recurrent neural network and differential autoencoder, characterized in that: include: The collected original trajectory data is segmented to obtain the segmented trajectory data, specifically: Obtain the vehicle's trajectory point location information and timestamp in the driving trajectory data; based on the vehicle's trajectory point location information and timestamp, sequence represents a trajectory in the trajectory data; Representation sequence Middle The timestamp and location information of each track point, is the length of the trajectory; Based on the trajectory segmentation rules, the original trajectory data To segment, use sequence represents a trajectory in the trajectory data; Representation sequence Middle The timestamp and location information of each track point, is the length of the trajectory; The trajectory segmentation rules are specifically as follows: When the spatial deviation of two adjacent trajectory points is greater than the set first threshold, Lagrange interpolation is performed through the surrounding trajectory points to form a complete trajectory data; or when the trajectory point does not move within the set time or the displacement deviation of the trajectory point does not reach the set second threshold, the trajectory is segmented to form a complete trajectory data; The segmentation of the trajectory is specifically as follows: Eliminate the trajectory points that have not moved within the set time, segment the non-moving displacement points, and obtain the segmented trajectory data; Divide the target map into several grids, discretize the segmented trajectory data, and obtain discretized trajectory data; Divide the discretized trajectory data to obtain training sets and test sets; Based on recurrent neural network and differential autoencoder, an abnormal trajectory detection model is constructed; Input the training set into the abnormal trajectory detection model for training to obtain the optimized abnormal trajectory detection model; Input the test set into the optimized abnormal trajectory detection model to obtain the abnormal score of the trajectory; Based on the anomaly score, determine whether the trajectory is abnormal.

2. The abnormal trajectory detection method of the recurrent neural network and differential autoencoder according to claim 1 is characterized in that: The target map is divided into a plurality of grids, and the segmented trajectory data is discretized to obtain the discretized trajectory data, specifically: The target map is divided into several grids. The trajectory points falling into the same grid have the same number to discretize the trajectory data points. represents a trajectory in the trajectory data; Representation sequence Middle The timestamp and location information of each track point, is the length of the trajectory.

3. The abnormal trajectory detection method of the recurrent neural network and differential autoencoder according to claim 2 is characterized in that: The abnormal trajectory detection model is constructed based on the recurrent neural network and the differential autoencoder, specifically: Step 4.1: Build the trajectory autoencoder model; including the encoder , probability distribution and generator ; Encoder For a recurrent neural network, the trajectory data in the training set As input, the latent representation of the output trajectory ; Probability distribution of the path is a probability distribution model based on Gaussian distribution, with the potential representation of the trajectory As input, output represents the path type in the latent space; is the number of species; generator For a recurrent neural network, the potential representation of the trajectory As input, output is the reconstructed trajectory ; Step 4.2: Build the Discriminator ; Discriminator for Layer fully connected neural network, the trajectory data in the training set and the trajectory of reconstruction As input, the output score matrix .

4. The abnormal trajectory detection method based on recurrent neural network and differential autoencoder according to claim 3 is characterized in that: The training set is input into the abnormal trajectory detection model for training to obtain the optimized abnormal trajectory detection model, specifically: Step 5.1: Variation of lower bound using trajectory information As the loss function, use the trajectory data in the training set , trajectory latent representation , the encoder of the user variational autoencoder is trained using stochastic gradient descent and generator Neural network parameters in ; Trajectory latent representation ; in, Represents the encoder to be trained and generator Neural network parameters in ; Representation trajectory The probability of belonging to the normal path; Step 5.2: Repeat step 6.1 to train the abnormal trajectory detection model until the loss function is lower than the set threshold.

5. The abnormal trajectory detection method of the recurrent neural network and differential autoencoder according to claim 4 is characterized in that: The test set is input into the optimized abnormal trajectory detection model to obtain the abnormal score of the trajectory, specifically: The trajectory data in the test set Input to the encoder part of the optimized abnormal trajectory detection model, and the output is the potential representation of the trajectory ; Based on the latent representation Compute the path in the latent space The probability of , and the anomaly score of the trajectory is calculated by the probability .

6. The abnormal trajectory detection method based on recurrent neural network and differential autoencoder according to claim 5 is characterized in that: Anomaly score The calculation method is: in, Represents the path latent representation in the latent vector space.

7. An abnormal trajectory detection system based on recurrent neural network and differential autoencoder, characterized in that: include: A segmentation module, wherein the segmentation module is used to segment the collected original trajectory data to obtain segmented trajectory data; The functionality of the segmentation module is implemented through the following methods: Obtain the vehicle's trajectory point location information and timestamp in the driving trajectory data; based on the vehicle's trajectory point location information and timestamp, sequence represents a trajectory in the trajectory data; Representation sequence Middle The timestamp and location information of each track point, is the length of the trajectory; Based on the trajectory segmentation rules, the original trajectory data To segment, use sequence represents a trajectory in the trajectory data; Representation sequence Middle The timestamp and location information of each track point, is the length of the trajectory; The trajectory segmentation rules are specifically as follows: When the spatial deviation of two adjacent trajectory points is greater than the set first threshold, Lagrange interpolation is performed through the surrounding trajectory points to form a complete trajectory data; or when the trajectory point does not move within the set time or the displacement deviation of the trajectory point does not reach the set second threshold, the trajectory is segmented to form a complete trajectory data; The segmentation of the trajectory is specifically as follows: Eliminate the trajectory points that have not moved within the set time, segment the non-moving displacement points, and obtain the segmented trajectory data; A first division module, wherein the first division module is used to divide the target map into a plurality of grids, discretize the segmented trajectory data, and obtain discretized trajectory data; A second partitioning module, the second partitioning module is used to partition the discretized trajectory data to obtain a training set and a test set; A building module, wherein the building module builds an abnormal trajectory detection model based on a recurrent neural network and a differential autoencoder; A training module, wherein the training module is used to input the training set into the abnormal trajectory detection model for training to obtain an optimized abnormal trajectory detection model; A testing module, wherein the testing module is used to input a test set into an optimized abnormal trajectory detection model to obtain an abnormal score of the trajectory; A judgment module is used to judge whether the trajectory is abnormal based on the abnormality score.

Images