Maneuvering target trajectory prediction method based on probability statistical compression

Abstract

The invention discloses a maneuvering target trajectory prediction method based on probability statistical compression, belongs to the technical field of autonomous navigation, traffic planning and remote sensing early warning, and aims to solve the problem that movement state changes are difficult to deal with in maneuvering target trajectory prediction. The main technology comprises the steps that a historical track data set is input, preprocessing is carried out, maneuvering information is extracted, and a processed data set is obtained; creating a state grid coordinate system; discretizing the state of each frame in the data set in a grid coordinate system, and storing corresponding maneuvering information and weight in limited state grid points to obtain a table T; extracting an optimal maneuvering decision from the table T to obtain a table T *; inputting the state of a to-be-predicted frame, querying a maneuvering decision of a neighborhood grid point in the table T *, and calculating an optimal maneuvering decision corresponding to the to-be-predicted state; applying the state of the frame to be predicted and the optimal maneuvering decision to a physical model, and predicting the state of the next moment; and the steps are repeatedly executed until the specified prediction step number, and multi-step prediction is realized.

Description

Technical Field

[0001] The invention belongs to the technical field of autonomous navigation, traffic planning and remote sensing early warning, and provides a method for predicting the trajectory of a maneuvering target based on probability statistics compression, which is used for predicting the trajectory of a maneuvering target. Background Art

[0002] Trajectory prediction refers to the task of predicting the future position, speed, direction and other state information of a target given its past or current trajectory. It is an important component in many fields and has a wide range of applications in autonomous navigation, traffic planning and remote sensing warning.

[0003] According to the variable selection method, trajectory prediction methods can be divided into three methods: local regression, global regression, and autoregression. Local regression usually uses time as the independent variable, and only uses a segment of trajectory data near the predicted point for regression each time. Local regression decomposes complex trajectories into several simple trajectories for processing, which reduces the complexity of modeling, but the long-term prediction effect is poor. At present, the commonly used local regression methods in trajectory prediction include local weighted linear regression, local weighted polynomial regression, etc.

[0004] The global regression model uses macro factors such as planned routes and environmental conditions as independent variables. Each prediction will output a complete trajectory, so there is no error accumulation problem in multi-step predictions. It has better long-term prediction capabilities, but it cannot make real-time predictions. In addition, it is difficult to express it with an explicit mathematical model. Therefore, compared with local regression, global regression places higher requirements on the structure and scale of the model. Currently, there are few prediction methods using global regression, and they are usually based on artificial neural networks.

[0005] Autoregression is the most widely used method in trajectory prediction. It uses the dependency (i.e., self-correlation) between the values ​​of the historical time series of the prediction target in different periods to establish a regression model for prediction. The difference between it and the above two regression methods is that the types of independent variables and dependent variables are the same. The autoregression model includes state estimation method and artificial neural network method.

[0006] The state estimation method establishes the state transfer equation based on the position, velocity, acceleration and other properties of the maneuvering target, and uses the observed state to continuously correct the estimation of the system state. However, this method has the following problems: it needs to continuously input the latest observed state during prediction, so it can only make single-step predictions; and its prediction of the maneuvering target is based on the state at the past moment, which makes it unable to capture the possible maneuvering changes of the maneuvering target in the future. Common state estimation methods include Kalman filtering, particle filtering and hidden Markov model.

[0007] The autoregressive model based on artificial neural network usually uses recurrent neural network as its basic structure. However, the prediction effect of the existing autoregressive model based on artificial neural network is not good, and its performance upper limit is only close to the state estimation method and local regression method. Summary of the invention

[0008] The purpose of the present invention is to provide a method for predicting the trajectory of a maneuvering target based on probability statistics compression, so as to solve the following problems in the prior art: the existing local regression and autoregression models are difficult to capture the changes in the future motion state of the maneuvering target, which makes them unsuitable for long-term prediction; although the global regression model has better long-term prediction performance in theory, the characteristic of generating a complete trajectory at one time makes it impossible to make real-time predictions based on the current state, and the model occupies a large space and the prediction speed is slow. The present invention has the following characteristics: fast prediction speed, small storage space occupation, and high accuracy of both short-term and long-term predictions.

[0009] In order to achieve the above object, the present invention adopts the following technical solution:

[0010] A method for predicting a maneuvering target trajectory based on probability statistics compression comprises the following steps:

[0011] Step 1: Input the historical trajectory dataset D, perform preprocessing and extract maneuver information to obtain the processed dataset D*;

[0012] Step 2, create a state grid coordinate system C;

[0013] Step 3: Dataset D * The status information of each frame in s n Discretize in the grid coordinate system C and convert the corresponding maneuver information u n and weights are stored in a finite state grid, thus obtaining table T;

[0014] Step 4: Extract the optimal maneuver decision from the maneuver information and weights in table T, thereby obtaining table T containing the optimal maneuver decision. * ;

[0015] Step 5: Input the state of the frame to be predicted, query the maneuver decision of the neighboring grid points in table T*, and calculate the optimal maneuver decision corresponding to the state to be predicted based on this;

[0016] Step 6: Apply the state of the frame to be predicted and the optimal maneuver decision obtained in step 5 to the physical model G to predict the state at the next moment;

[0017] Step 7: Repeat steps 5 and 6 until the specified number of prediction steps is reached to achieve multi-step prediction.

[0018] Further, the data set D processed by step 1 is* Each frame contains status information s n and mobility information n , which is specifically expressed by the following formula:

[0019] D * [n] = {s n , v n}

[0020] State information refers to: known physical information that can reflect the current motion state of the target, including but not limited to position, speed and direction angle;

[0021] Maneuvering information refers to physical information that can reflect the current motion trend of the target and will have an indirect or direct impact on future state information, including but not limited to acceleration and angular velocity.

[0022] Further, the specific steps of step 2 are as follows:

[0023] Step 2.1, specify the grid coordinate range and resolution (grid size) of each state variable, the state variables include position, velocity, and direction angle;

[0024] Step 2.2: For each state variable, within the value range, create an arithmetic progression with the resolution as the step size, that is, the state grid coordinate axis. The grid coordinate axes of all state variables determine the state grid coordinate system C.

[0025] Furthermore, T in step 3 is a state grid point set S - is the index range, and the subtable T′ is a secondary nested table of values, which satisfies the following mapping relationship:

[0026] T:S - →T′

[0027] Where T' is a table with the maneuver information set U as the index range and the weight set W as the value range, which satisfies the following mapping relationship:

[0028] T′: U→W

[0029] Further, the specific steps of step 3 are as follows:

[0030] Step 3.1: Input dataset D * In the nth frame {s n ,u n}, find the status information s n The position in the state grid coordinate system, with this position as the center, takes the state grid points in its neighborhood as the set

[0031] Step 3.2: Calculate the probability contribution weight of the current frame to each state grid point in its neighborhood, and add the weight and maneuver information to each state grid point. The specific steps are as follows:

[0032] Step 3.2.1: For the collection The mth state grid point in With the specified kernel function K 1 Calculate n The weight w of its probability contribution n，m , which is specifically expressed by the following formula:

[0033]

[0034] Step 3.2.2: If the state grid point is not included in the table T, add the grid point to the table, and then add the maneuver information u n and weight w n，m Append to the sub-table T′ corresponding to the grid point;

[0035] Step 3.2.3: Repeat steps 3.2.1 and 3.2.2 until the collection is traversed.

[0036] Step 3.4: Repeat steps 3.1 and 3.2 until the dataset D is traversed. * .

[0037] Furthermore, T in step 4 * is a state grid set S - is the index range, the optimal maneuver information set U * is a table of value ranges, that is, it satisfies the following mapping relationship:

[0038] T * :S - →U *

[0039] Further, the specific steps of step 4 are as follows:

[0040] For the implementation of the present invention, step 4.1 provides the following three implementation routes:

[0041] Route 1:

[0042] For the mth state grid point in table T The corresponding subtable T′ contains k maneuver information and their corresponding weights, which are set U and set W respectively, to specify the kernel function K 2 The kernel density estimation method is used to calculate the probability density function of the maneuver, and the maneuver with the maximum probability density is taken as the optimal maneuver corresponding to the current state grid point. The specific formula is as follows:

[0043]

[0044]

[0045] Where p represents the probability density function of the maneuver, x is the independent variable of the probability density function, h is the specified bandwidth parameter, and u i is the i-th maneuver information in table T′, w i for u i The corresponding weights;

[0046] Route 2:

[0047] For the mth state grid point in table T The corresponding subtable T′ contains several maneuver information and their corresponding weights. The histogram statistics of T′ are performed at the specified resolution h, and the midpoint of the maneuver interval with the maximum cumulative weight is taken as the grid optimal maneuver

[0048] Route 3:

[0049] For the mth state grid point in table T The corresponding subtable T′ contains several maneuver information and their corresponding weights. T′ is regarded as a Gaussian mixture model. The expected maximum algorithm is used to estimate multiple means of the mixed distribution of maneuver information, and the mean with the largest probability is taken as the optimal maneuver of the grid.

[0050] Step 4.2: Repeat step 4.1 until table T is traversed and table T is obtained. * .

[0051] Further, the specific steps of step 5 are as follows:

[0052] Step 5.1: Find the position of the state s of the frame to be predicted in the state grid coordinate system, take the state grid points in its neighborhood as the center, and take the state grid points in its neighborhood as the set

[0053] Step 5.2: The element is the index in table T * Query in sequence to obtain the grid optimal maneuver set

[0054] Step 5.3: For The nth state grid point in With the specified kernel function K 3 Computational Grid Optimal Maneuver The probability contribution weight The specific formula is as follows:

[0055]

[0056] Step 5.4: Calculate the weight The weighted average of the current state is used to obtain the optimal maneuver decision corresponding to the current state. The specific formula is as follows:

[0057]

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] First, the prediction model used in the present invention can extract the maneuvering decisions of maneuvering targets in different states from a large number of historical trajectories, overcoming the problem of the prior art that it is difficult to accurately predict when the motion state changes;

[0060] 2. The present invention compresses data through probabilistic statistical methods and uses sparse storage methods, which can greatly reduce the space occupied by the prediction model.

[0061] 3. The present invention completes most of the calculations in the stage of establishing the prediction model, compressing the information attached to a large amount of historical data into information in a small number of state grid points, so that it has a very fast calculation speed in the prediction stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] 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.

[0063] Figure 1 This is a flow chart of Embodiment 1 of the present invention;

[0064] Figure 2 This is an effect diagram and a partial enlarged diagram of a single-step prediction of a certain trajectory in accordance with an embodiment of the present invention.

[0065] Figure 3 This is a diagram showing the effect of long-term prediction of a certain trajectory in accordance with an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments:

[0067] Embodiment 1

[0068] A method for predicting a maneuvering target trajectory based on probability statistics compression comprises the following steps:

[0069] Step 1: Input the historical trajectory dataset D, perform preprocessing and extract acceleration and angular velocity to obtain the processed dataset D * ;

[0070] Step 2: Create a state grid coordinate system using longitude, latitude, speed and direction angle as state variables;

[0071] Step 3: Use the kernel density estimation method to transform the data set D * The state s of each frame in n Discretize in the grid coordinate system C, and store the corresponding acceleration, angular velocity and weight into a finite state grid, thus obtaining table T a and Table T ω ;

[0072] Step 4: Use the Gaussian mixture model estimation method to obtain the a and Table T ω The optimal acceleration and angular velocity strategy is extracted from the acceleration, angular velocity and weight information in Table and table

[0073] Step 5: Enter the status information of the frame to be predicted. Query the grid optimal maneuver decision in , and calculate the optimal maneuver information of the frame state to be predicted based on this;

[0074] Step 6: Apply the state information of the frame to be predicted and the optimal maneuver information obtained in step 5 to a linear approximation turning rate and acceleration (TRA) model to predict the state at the next moment;

[0075] Step 7: Repeat steps 5 and 6 until the specified number of prediction steps is reached to achieve multi-step prediction.

[0076] Further, the specific steps of step 1 are as follows:

[0077] Step 1.1: Input a track from the historical track data set and reconstruct the time based on the latitude, longitude and speed data. The specific steps are as follows:

[0078] Step 1.1.1. Use the haversine formula to calculate the displacement d between two adjacent frames (the nth frame and the n+1th frame) n , which is specifically expressed by the following formula:

[0079]

[0080] Where Lon and Lat represent longitude and latitude respectively, and R is the radius of the earth, which is approximately 6378137 meters.

[0081] Step 1.1.2, displacement dn Divide the speed v of the previous frame n The interval time between two frames can be obtained, and the time t after the current frame reconstruction can be obtained by accumulating the interval time of all previous frames. n , which is specifically expressed by the following formula:

[0082]

[0083] Step 1.1.3: Repeat steps 1.1 and 1.2 until the entire trajectory is traversed.

[0084] Step 1.2: Resample the trajectory processed in step 1.1 at a fixed time interval. The specific steps are as follows:

[0085] Step 1.2.1, create an arithmetic sequence with the time interval Δt=10s and the time range of the current trajectory as interpolation points;

[0086] Step 1.2.2: For a state variable in the trajectory data, use the cubic spline interpolation method to fit the reconstructed time to obtain a cubic spline interpolation equation set and its coefficients;

[0087] Step 1.2.3, substitute the interpolation points obtained in step 2.2 into the equations obtained in step 2.1, and use the corresponding coefficients to find the function value as the resampling result of the variable;

[0088] Step 1.2.4: Repeat steps 2.2 and 2.3 until all types of variables are traversed.

[0089] Step 1.3: Extract the acceleration and angular velocity from the trajectory obtained in step 1.2. The specific steps are as follows:

[0090] Step 1.3.1: Use forward difference to process the velocity and get the velocity change Δv n , divided by the resampling time interval Δt to obtain the acceleration a n ;

[0091] Step 1.3.2: Use forward difference to process the velocity direction to obtain the velocity direction change Δθ n , and then calculate the angular velocity ω according to the following formula n :

[0092]

[0093] Step 1.3.3: Repeat steps 3.1 to 3.4 until the entire trajectory is traversed.

[0094] Step 1.4: Repeat steps 1.1 to 1.3 until the historical trajectory dataset is traversed.

[0095] Further, the data set D processed by step 1 is * Each frame contains status information: longitude lon n , latitude lat n , speed v n 、Direction angle θ n ; and maneuver information: acceleration a n and angular velocity ω n , which is specifically expressed by the following formula:

[0096] D * [n] = {lon n ,lat n , v n ,θ n , a n ,ω n}

[0097] Further, the specific steps of step 2 are as follows:

[0098] Step 2.1. Specify the grid coordinate range and resolution of longitude to be [115.5°, 121.5°] and 0.01° respectively; the grid coordinate range and resolution of latitude to be [30.5°, 40.5°] and 0.01° respectively; the grid coordinate range and resolution of velocity to be [0m / s, 350m / s] and 1m / s respectively; the grid coordinate range and resolution of azimuth to be [0°, 360°] and 1° respectively.

[0099] Step 2.2: For each state variable, create an arithmetic progression within the range of values ​​with the resolution as the step size, i.e., the grid coordinate axis. The grid coordinate axes of the state variables (longitude, latitude, and velocity) that are highly correlated with acceleration determine the grid coordinate system C. 1 The grid coordinate axes of the state variables (longitude, latitude and azimuth) that are highly correlated with the angular velocity determine the grid coordinate system C 2 .

[0100] Further, the specific steps of step 3 are as follows:

[0101] Step 3.1: Input the nth frame {lon n ,lat n , v n ,θ n , a n ,ω n}, find its acceleration related state s n = {lon n ,lat n , v n}In the grid coordinate system C 1The position in the window with a width of 4 is taken as the set, with the position as the center.

[0102] Step 3.2: Calculate the probability weight contributed by each state grid point in the neighborhood of the current frame, and compress the weight and acceleration into each state grid point. The specific steps are as follows:

[0103] Step 3.2.1: For the mth state grid point in the neighborhood Calculate the probability contribution weight w of the current frame using the Gaussian kernel function n，m , which is specifically expressed by the following formula:

[0104]

[0105] Where ∑ is the covariance matrix of the state, and its specific values ​​are as follows:

[0106]

[0107] Step 3.2.2, if table T a If the state grid point is not included in the table, add the grid point to the table, and then add the maneuver information u n and weight w n，m Append to the table T corresponding to the grid point a 'middle.

[0108] Step 3.2.3: Repeat steps 3.2.1 and 3.2.2 until the collection is traversed.

[0109] Step 3.4: Repeat steps 3.1 and 3.2 until the historical trajectory dataset is traversed.

[0110] Step 3.5: Compress the angular velocity in the data set into a table T indexed by longitude, latitude and direction angle. ω The specific process is the same as the above process.

[0111] Further, the specific steps of step 4 are as follows:

[0112] Step 4.1: For the acceleration set A and weight set W of a certain state grid point in the grid acceleration weight table, regard them as Gaussian mixture models, use the expected maximum algorithm to estimate multiple means of the acceleration mixture distribution, and take the mean with the largest probability as the grid optimal acceleration a. * ;

[0113] The acceleration weight list data is overwritten with the optimal maneuver data.

[0114] Step 4.2: Repeat step 4.1 until the grid acceleration weight table is traversed and the grid optimal acceleration table is obtained.

[0115] Step 4.3: Use the same method as steps 4.1 and 4.2 to process the grid angular velocity weight table to obtain the grid optimal angular velocity table.

[0116] Further, the specific steps of step 5 are as follows:

[0117] Step 5.1, respectively find the state s = {lon, lat, v, θ} of the frame to be predicted in the coordinate system C 1 and coordinate system C 2 The position in the middle, taking the position as the center, takes the state grid points in its neighborhood as the set

[0118] Step 5.2: Use S - The elements of and table Query the optimal maneuver decision in the neighborhood grid point to obtain the optimal maneuver set U *

[0119] Step 5.3: For the nth state grid point in the neighborhood Computing the Optimal Maneuver of a Grid Using Gaussian Kernel Function The probability contribution weight The specific formula is as follows:

[0120]

[0121] Step 5.4: Calculate the weight The weighted average of the current state is used to obtain the optimal maneuver decision corresponding to the current state. The specific formula is as follows:

[0122]

[0123] Further, the step 6 is specifically expressed by the formula as follows:

[0124]

[0125] Where s′ represents the predicted state after Δt time.

[0126] In summary, the method for predicting maneuvering target trajectory based on probability statistics compression proposed in the present invention has the following technical effects and advantages:

[0127] 1. Historical trajectory data preprocessing and maneuver information extraction: Through time reconstruction, resampling and extraction of maneuver information, historical trajectory data can be effectively processed to provide an accurate data basis for subsequent predictions.

[0128] 2. Probabilistic statistical compression and state grid establishment: The maneuver information is compressed twice into a table indexed by the state grid to reduce storage space requirements.

[0129] 3. Fast query and prediction: Most of the calculations are completed in the data compression stage. When predicting, the maneuvering information of neighboring state grid points can be quickly queried, and predictions can be made in combination with physical models, which improves the prediction speed.

[0130] 4. Solve the problem of motion state change: The present invention can extract the maneuvering decisions of maneuvering targets in different states from a large number of historical trajectories, effectively solving the problem that the prior art is difficult to accurately predict when the motion state changes.

[0131] 5. Multi-step prediction and high accuracy: By repeating the query and prediction steps, multi-step prediction is achieved while maintaining high prediction accuracy.

[0132] In summary, the present invention provides an effective method in the field of maneuvering target trajectory prediction, especially has significant advantages in processing the change of the motion state of the maneuvering target, and has the characteristics of fast prediction speed and small storage space occupation.

Claims

1. A method for predicting maneuvering target trajectory based on probability statistics compression, characterized in that: The following steps are involved: Step 1: Input the historical trajectory dataset D, perform preprocessing and extract maneuver information to obtain the processed dataset D * ; Step 2, create a state grid coordinate system C; Step 3: Dataset D * The status information of each frame in s n Discretize in the grid coordinate system C and convert the corresponding maneuver information u n and weights are stored in a finite state grid, thus obtaining table T; Step 4: Extract the optimal maneuver decision from the maneuver information and weights in table T, thereby obtaining table T containing the optimal maneuver decision. * ; Step 5: Enter the state of the frame to be predicted in Table T * Query the maneuvering decision of the neighborhood grid points in , and on this basis calculate the optimal maneuvering decision corresponding to the state to be predicted; Step 6: Apply the state of the frame to be predicted and the optimal maneuver decision obtained in step 5 to the physical model G to predict the state at the next moment; Step 7: Repeat steps 5 and 6 until the specified number of prediction steps is reached to achieve multi-step prediction.

2. The method for predicting maneuvering target trajectory based on probability statistics compression according to claim 1, characterized in that: The dataset D processed by step 1 * Each frame contains status information s n and mobility information n , which is specifically expressed by the following formula: D * [n]={s n ,u n } State information refers to: known physical information that can reflect the current motion state of the target, including position, speed and direction angle; Maneuvering information refers to physical information that can reflect the current motion trend of the target and will have an indirect or direct impact on future state information, including acceleration and angular velocity.

3. The method for predicting maneuvering target trajectory based on probability statistics compression according to claim 1, characterized in that: The specific steps of step 2 are as follows: Step 2.1, specify the grid coordinate range and resolution of each state variable, the state variables include position, velocity, and direction angle; Step 2.2: For each state variable, within the value range, create an arithmetic progression with the resolution as the step size, that is, the state grid coordinate axis. The grid coordinate axes of all state variables determine the state grid coordinate system C.

4. The method for predicting maneuvering target trajectory based on probability statistics compression according to claim 1, characterized in that: The T in step 3 is a state grid point set S - is the index range, and the subtable T′ is a secondary nested table of values, which satisfies the following mapping relationship: T′:S - →T′ Where T′ is a table with the maneuver information set U as the index range and the weight set W as the value range, which satisfies the following mapping relationship: T′:U→W.

5. The method for predicting maneuvering target trajectory based on probability statistics compression according to claim 1, characterized in that: The specific steps of step 3 are as follows: Step 3.1: Input dataset D * In the nth frame {s n ,u n }, find the status information s n The position in the state grid coordinate system, with this position as the center, takes the state grid points in its neighborhood as the set Step 3.2: Calculate the probability contribution weight of the current frame to each state grid point in its neighborhood, and add the weight and maneuver information to each state grid point. The specific steps are as follows: Step 3.2.1: For the collection The mth state grid point in Calculate s with the specified kernel function K1 n The weight w of its probability contribution n，m , which is specifically expressed by the following formula: Step 3.2.2: If the state grid point is not included in the table T, add the grid point to the table, and then add the maneuver information u n and weight w n，m Append to the sub-table T′ corresponding to the grid point; Step 3.2.3: Repeat steps 3.2.1 and 3.2.2 until the collection is traversed. Step 3.4: Repeat steps 3.1 and 3.2 until the dataset D is traversed. * .

6. The method for predicting maneuvering target trajectory based on probability statistics compression according to claim 1, characterized in that: In step 4, T * is a state grid set S - is the index range, the optimal maneuver information set U * is a table of value ranges, that is, it satisfies the following mapping relationship: T * :S - →U * 。 7. The method for predicting maneuvering target trajectory based on probability statistics compression according to claim 1, characterized in that: The specific steps of step 4 are as follows: Step 4.1: For the mth state grid point in table T The corresponding subtable T′ contains k maneuver information and their corresponding weights, which are set U and set W respectively. The probability density function of the maneuver is calculated using the kernel density estimation method with the specified kernel function K2, and the maneuver with the maximum probability density is taken as the optimal maneuver corresponding to the current state grid point. The specific formula is as follows: Where p represents the probability density function of the maneuver, x is the independent variable of the probability density function, h is the specified bandwidth parameter, and u i is the i-th maneuver information in table T′, w i for u i The corresponding weights; Step 4.2: Repeat step 4.1 until all state grid points in table T are traversed to obtain table T * .

8. The method for predicting maneuvering target trajectory based on probability statistics compression according to claim 7, characterized in that: The specific steps of step 4 are as follows: Step 4.1: For the mth state grid point in table T The corresponding subtable T′ contains several maneuver information and their corresponding weights. T′ is regarded as a Gaussian mixture model. The expected maximum algorithm is used to estimate multiple means of the mixed distribution of maneuver information, and the mean with the largest probability is taken as the optimal maneuver corresponding to the current state grid point. Step 4.2: Repeat step 4.1 until all state grid points in table T are traversed to obtain table T * .

9. The method for predicting maneuvering target trajectory based on probability statistics compression according to claim 1, characterized in that: The specific steps of step 4 are as follows: Step 4.1: For the mth state grid point in table T The corresponding subtable T′ contains several maneuver information and their corresponding weights. The histogram statistics of T′ are performed at the specified resolution h, and the midpoint of the maneuver interval with the maximum cumulative weight is taken as the optimal maneuver corresponding to the current state grid point. Step 4.2: Repeat step 4.1 until all state grid points in table T are traversed to obtain table T * .

10. The method for predicting maneuvering target trajectory based on probability statistics compression according to claim 1, characterized in that: The specific steps of step 5 are as follows: Step 5.1: Find the position of the state s of the frame to be predicted in the state grid coordinate system, take the state grid points in its neighborhood as the center, and take the state grid points in its neighborhood as the set Step 5.2: The element is the index in table T * Query in sequence to obtain the grid optimal maneuver set Step 5.3: For The nth state grid point in Compute the optimal maneuver of the grid with the specified kernel function K3 The probability contribution weight The specific formula is as follows: Step 5.4: The weighted average of the optimal maneuver set of the weighted calculation grid is used to obtain the optimal maneuver decision corresponding to the current state The specific formula is as follows:

Images