Path prediction method and device based on time sequence convolutional network

Abstract

The invention provides a path prediction method and device based on a time sequence convolutional network, and relates to the technical field of automatic control, and the method comprises the steps: obtaining obstacle information, and target position coordinate information, current position coordinate information and historical position coordinate information of a moving object; generating first state feature information of the current position according to the target position coordinate information, the current position coordinate information, the historical position coordinate information and the obstacle information; generating second state feature information of a plurality of historical positions according to the target position coordinate information, the historical position coordinate information and the obstacle information; constructing a feature information sequence in combination with the first state feature information and the second state feature information; and inputting the feature information sequence into a pre-generated path prediction model to obtain the coordinate information of the next moving position of the moving object. According to the invention, efficient, smooth and safe obstacle avoidance path prediction of the next position point of the moving object is realized, and the real-time performance and reliability of path planning are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic control, and in particular to a path prediction method and device based on a time sequence convolution network. BACKGROUND

[0002] Path planning is a key technology for determining an optimal or relatively optimal motion path for a moving object (such as a robot, an autonomous vehicle, etc.) from a starting point to a target point in a dynamic or complex environment. The core goal is to guide the moving object to avoid obstacles, reach the destination smoothly and efficiently under the premise of meeting environmental constraints, safety requirements and kinematic limitations.

[0003] The traditional path planning process usually includes the following steps: first, the position information of the moving object in the environment is obtained in real time through a sensor to determine the current position point as the starting basis for path planning; then, based on a pre-constructed environmental model (such as map information, obstacle distribution, etc.) and a path planning algorithm, a plurality of candidate prediction points are generated by searching for possible next moving positions around the current position; subsequently, the rationality and feasibility of these prediction points are evaluated from multiple dimensions, both checking whether the candidate points will collide with obstacles and evaluating their directional relationship with the target point, and the candidate points that are oriented towards the target and avoid obstacles are preferentially selected as the next position points; once a suitable point is selected, the moving object will move to that point according to the control mechanism and restart the search process to iteratively generate subsequent path points. Finally, by sequentially connecting all the determined path points, a complete path trajectory is formed for the moving object to execute.

[0004] However, the existing path planning methods still have limitations in handling complex scenarios, responding to changes in real time, and fusing historical trajectory information for prediction. For example, traditional algorithms are mostly based on rule construction of heuristic functions, which are difficult to effectively utilize long-term dependency information in time series data, resulting in insufficient prediction accuracy and path smoothness. SUMMARY

[0005] To solve the problems in the prior art, the embodiments of the present application provide a path prediction method and device based on a time sequence convolution network, which can solve the problems in the prior art.

[0006] In a first aspect, the present application provides a path prediction method based on a time sequence convolution network, comprising:

[0007] obstacle information and position information of the moving object; the position information includes target position coordinate information, current position coordinate information and historical position coordinate information;

[0008] generating first state feature information of the current position according to the target position coordinate information, the current position coordinate information, the historical position coordinate information and the obstacle information;

[0009] generate second state feature information of a plurality of historical positions according to the target position coordinate information, the historical position coordinate information and the obstacle information;

[0010] construct a feature information sequence in combination with the first state feature information and the second state feature information;

[0011] input the feature information sequence into a path prediction model generated in advance to obtain next moving position coordinate information of the moving object.

[0012] Further, the historical position coordinate information includes last moving position coordinate information of a current position of the moving object; and the first state feature information of the current position is generated according to the target position coordinate information, the current position coordinate information, the historical position coordinate information and the obstacle information, including:

[0013] calculate a first average Euclidean distance from the current position to the obstacle according to the current position coordinate information and the obstacle information;

[0014] calculate a first Euclidean distance from the current position to the target position according to the current position coordinate information and the target position coordinate information;

[0015] determine a first included angle of a heading of the current position and a line connecting the target point according to the current position coordinate information, the last moving position coordinate information and the target position coordinate information;

[0016] construct the first state feature information by using the current position coordinate information, the first average Euclidean distance, the first Euclidean distance and the first included angle.

[0017] Further, the second state feature information of a plurality of historical positions is generated according to the target position coordinate information, the historical position coordinate information and the obstacle information, including:

[0018] calculate a second average Euclidean distance from each historical position to the obstacle according to a plurality of the historical position coordinate information and the obstacle information;

[0019] calculate a second Euclidean distance from each historical position to the target position according to a plurality of the historical position coordinate information and the target position coordinate information;

[0020] determine a second included angle of a heading of each historical position and a line connecting the target point according to a plurality of the historical position coordinate information and the target position coordinate information;

[0021] construct a plurality of the second state feature information by using the historical position coordinate information, the second average Euclidean distance, the second Euclidean distance and the second included angle.

[0022] Furthermore, the step of pre-generating the path prediction model includes:

[0023] An initial model is constructed using a temporal convolutional network structure, wherein the initial model includes a fully connected layer, a temporal convolutional network module, a feature shaping module, and a multi-layer perceptron module;

[0024] The initial model is trained using pre-generated training samples to predict the next moving position coordinate information of the moving object;

[0025] The prediction error is calculated using a pre-built composite loss model, and the parameters of the trained initial model are optimized through back propagation until the model converges, thereby obtaining the path prediction model.

[0026] Furthermore, the step of pre-building the composite loss model includes:

[0027] generating a first superposition control item according to a third Euclidean distance from the predicted position to the target position and a first Euclidean distance from the current position to the target position;

[0028] generating a second superposition control item according to a third angle between a line connecting the current position and the predicted position and a line connecting the current position and the target position;

[0029] generating a third superposition control item according to a third average Euclidean distance from the predicted position to the obstacle;

[0030] The composite loss model is constructed according to the pre-constructed initial loss model, the first superposition control item, the second superposition control item and the third superposition control item.

[0031] In a second aspect, the present application provides a path prediction device based on a temporal convolutional network, comprising:

[0032] An information acquisition unit, configured to acquire obstacle information and position information of a moving object; the position information includes target position coordinate information, current position coordinate information, and historical position coordinate information;

[0033] a current state characteristic information generating unit, configured to generate first state characteristic information of the current position based on the target position coordinate information, the current position coordinate information, the historical position coordinate information, and the obstacle information;

[0034] a historical state characteristic information generating unit, configured to generate second state characteristic information of a plurality of historical positions according to the target position coordinate information, the historical position coordinate information, and the obstacle information;

[0035] a feature information sequence construction unit, configured to construct a feature information sequence by combining the first state feature information and the second state feature information;

[0036] a path prediction unit, configured to input the feature information sequence into a pre-generated path prediction model to obtain next moving position coordinate information of the moving object.

[0037] Further, the historical position coordinate information includes last moving position coordinate information of a current position of the moving object; and the current state feature information generation unit includes:

[0038] an obstacle distance calculation module, configured to calculate a first average Euclidean distance from the current position to an obstacle according to the current position coordinate information and the obstacle information;

[0039] a target position distance calculation module, configured to calculate a first Euclidean distance from the current position to a target position according to the current position coordinate information and the target position coordinate information;

[0040] an included angle calculation module, configured to determine a first included angle between a heading and a line connecting the target point according to the current position coordinate information, the last moving position coordinate information and the target position coordinate information;

[0041] a current state feature information generation module, configured to construct the first state feature information by using the current position coordinate information, the first average Euclidean distance, the first Euclidean distance and the first included angle.

[0042] Further, the historical state feature information generation unit includes:

[0043] a historical obstacle distance calculation module, configured to calculate a second average Euclidean distance from each historical position to the obstacle according to the plurality of historical position coordinate information and the obstacle information;

[0044] a historical target position distance calculation module, configured to calculate a second Euclidean distance from each historical position to the target position according to the plurality of historical position coordinate information and the target position coordinate information;

[0045] a historical included angle calculation module, configured to determine a second included angle between a heading of each historical position and a line connecting the target point according to the plurality of historical position coordinate information and the target position coordinate information;

[0046] a historical state feature information generation module, configured to construct the plurality of second state feature information by using the historical position coordinate information, the second average Euclidean distance, the second Euclidean distance and the second included angle.

[0047] Further, the method further includes:

[0048] The model construction unit is configured to construct an initial model by using a time sequence convolution network structure, the initial model comprising a full connection layer, a time sequence convolution network module, a feature shaping module and a multi-layer perception module.

[0049] The model training unit is configured to train the initial model by using the pre-generated training samples to predict the next moving position coordinate information of the moving object.

[0050] The model optimization unit is configured to calculate a prediction error by using a pre-constructed compound loss model, and optimize the parameters of the trained initial model by back propagation until the model converges, to obtain the path prediction model.

[0051] Further, the method further comprises:

[0052] The first superimposed control item generation unit is configured to generate a first superimposed control item according to a third Euclidean distance from the predicted position to the target position and a first Euclidean distance from the current position to the target position.

[0053] The second superimposed control item generation unit is configured to generate a second superimposed control item according to a third included angle between a line connecting the current position and the predicted position and a line connecting the current position and the target position.

[0054] The third superimposed control item generation unit is configured to generate a third superimposed control item according to a third average Euclidean distance from the predicted position to the obstacle.

[0055] The compound loss model construction unit is configured to construct the compound loss model according to the pre-constructed initial loss model, the first superimposed control item, the second superimposed control item and the third superimposed control item.

[0056] In a third aspect, the present application provides a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the path prediction method based on the time sequence convolution network according to any one of the above embodiments.

[0057] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to implement the path prediction method based on the time sequence convolution network according to any one of the above embodiments.

[0058] In a fifth aspect, the present application provides a computer program product, which comprises a computer program, wherein the computer program is executed by a processor to implement the path prediction method based on the time sequence convolution network according to any one of the above embodiments.

[0059] The application provides a path prediction method and device based on a time sequence convolution network, which obtains obstacle information and position information of a moving object; the position information comprises target position coordinate information, current position coordinate information and historical position coordinate information; first state feature information of a current position is generated according to the target position coordinate information, the current position coordinate information, the historical position coordinate information and the obstacle information; second state feature information of a plurality of historical positions is generated according to the target position coordinate information, the historical position coordinate information and the obstacle information; a feature information sequence is constructed by combining the first state feature information and the second state feature information; the feature information sequence is input into a pre-generated path prediction model to obtain next moving position coordinate information of the moving object, thereby realizing efficient, smooth and obstacle-avoiding safe path prediction of a next position point of the moving object and improving real-time performance and reliability of path planning.

[0060] The application provides a path prediction method and device based on a time sequence convolution network, which obtains obstacle information and position information of a moving object; the position information comprises target position coordinate information, current position coordinate information and historical position coordinate information; first state feature information of a current position is generated according to the target position coordinate information, the current position coordinate information, the historical position coordinate information and the obstacle information; second state feature information of a plurality of historical positions is generated according to the target position coordinate information, the historical position coordinate information and the obstacle information; a feature information sequence is constructed by combining the first state feature information and the second state feature information; the feature information sequence is input into a pre-generated path prediction model to obtain next moving position coordinate information of the moving object, thereby realizing efficient, smooth and obstacle-avoiding safe path prediction of a next position point of the moving object and improving real-time performance and reliability of path planning. BRIEF DESCRIPTION OF DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, brief introductions will be given to the drawings needed to be used in the embodiments or prior art descriptions. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0062] Figure 1 is a flowchart of a path prediction method based on a time sequence convolution network provided by an embodiment of the present application;

[0063] Figure 2is a flowchart of a path prediction method based on a time sequence convolution network provided by an embodiment of the present application;

[0064] Figure 3 is a flowchart of a path prediction method based on a time sequence convolution network provided by an embodiment of the present application;

[0065] Figure 4 is a flowchart of a path prediction method based on a time sequence convolution network provided by an embodiment of the present application;

[0066] Figure 5 is a flowchart of a path prediction method based on a time sequence convolution network provided by an embodiment of the present application;

[0067] Figure 6 is a structural diagram of a path prediction device based on a time sequence convolution network provided by an embodiment of the present application;

[0068] Figure 7 is a structural diagram of a path prediction device based on a time sequence convolution network provided by an embodiment of the present application;

[0069] Figure 8 is a structural diagram of a path prediction device based on a time sequence convolution network provided by an embodiment of the present application;

[0070] Figure 9 is a structural diagram of a path prediction device based on a time sequence convolution network provided by an embodiment of the present application;

[0071] Figure 10 is a structural diagram of a path prediction device based on a time sequence convolution network provided by an embodiment of the present application;

[0072] Figure 11 is a schematic block diagram of a system structure of an electronic device provided by an embodiment of the present application;

[0073] Figure 12 is a path planning diagram provided by an embodiment of the present application;

[0074] Figure 13 is a schematic diagram of an overall flow of a heuristic path prediction method based on a time sequence convolution network provided by an embodiment of the present application;

[0075] Figure 14 is an architecture diagram of a heuristic path prediction model PathPreTCN provided by an embodiment of the present application. DETAILED DESCRIPTION

[0076] To make the purposes, technical solutions, and advantages of the embodiments of the present application clearer, further detailed descriptions of the embodiments of the present application are given below with reference to the drawings. Here, the illustrative embodiments of the present application and their descriptions are used to explain the present application but are not used as limitations of the present application. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other in any manner without conflicts.

[0077] To realize intelligent prediction of the path of a moving object, the present application proposes a path prediction method based on a temporal convolutional network, and proposes a heuristic path prediction model PathPreTCN based on a temporal convolutional network (TCN), which predicts information of a next position point based on information of a current position point and information of historical positions, combines spatial information (position) and time domain information (time sequence) of the information, and realizes optimized configuration of the next path position, while meeting the form safety of a moving vehicle, and also considering the smoothness of path changes. Meanwhile, in combination with information of historical position points, the search time of the next position point is reduced, and the efficiency of path prediction is improved.

[0078] Heuristic prediction is an algorithm strategy that uses additional information or empirical knowledge (i.e., heuristic information) to guide the prediction process to improve the search efficiency and quality. In the path planning method proposed in the present method, the heuristic information is the information of the current position and the historical position points (historical data information). The historical data plays an important role in predicting the next position point. It can reflect the past motion patterns, environmental change rules, and other information of the moving object. In the scenario of repeated task execution of a robot, the historical operation data can show which paths are more efficient and which regions are prone to unexpected conditions, and the algorithm can select and adjust the prediction points according to this, so that the selection of the prediction points is more in line with the actual situation and task requirements.

[0079] Path planning needs to consider multiple factors: first, the distance is short, and the path length from the starting point to the target point of the moving object is shortened as much as possible, so as to reduce the motion time and resource consumption. For example, in logistics distribution, a short path can reduce transportation costs and improve distribution efficiency. Second, the smoothness, the planned path should avoid sudden turns and mutations, and maintain smoothness, which is very important for stable operation of the moving object. For example, when an autonomous vehicle is driving, a smooth path can improve the ride comfort and reduce the wear of vehicle mechanical parts. Third, a safe distance from obstacles is maintained to ensure that the moving object has enough safety interval from pedestrians, other vehicles, fixed facilities, and other obstacles throughout the journey, effectively avoiding collision accidents, and ensuring the safety of itself and surrounding objects.

[0080] The specific implementation process of the path prediction method based on the temporal convolutional network provided by the embodiments of the present application is described below taking a server as an execution subject.

[0081] Figure 1 This is a flow chart of a path prediction method based on a temporal convolutional network provided by an embodiment of the present application. Figure 1 As shown, the path prediction method based on the temporal convolutional network provided by this application includes:

[0082] S101: Obtain obstacle information and position information of a moving object; the position information includes target position coordinate information, current position coordinate information, and historical position coordinate information;

[0083] S102: Generate first state feature information of the current position according to the target position coordinate information, the current position coordinate information, the historical position coordinate information, and the obstacle information;

[0084] S103: generating second state feature information of a plurality of historical positions according to the target position coordinate information, the historical position coordinate information, and the obstacle information;

[0085] S104: Constructing a characteristic information sequence by combining the first state characteristic information and the second state characteristic information;

[0086] S105: Input the feature information sequence into a pre-generated path prediction model to obtain the next moving position coordinate information of the moving object.

[0087] from Figure 1 As can be seen from the shown process, the present application provides a path prediction method based on a temporal convolutional network, which obtains obstacle information and position information of a moving object; the position information includes target position coordinate information, current position coordinate information and historical position coordinate information; first state feature information of the current position is generated according to the target position coordinate information, the current position coordinate information, the historical position coordinate information and the obstacle information; second state feature information of multiple historical positions is generated according to the target position coordinate information, the historical position coordinate information and the obstacle information; a feature information sequence is constructed by combining the first state feature information and the second state feature information; the feature information sequence is input into a pre-generated path prediction model to obtain the next moving position coordinate information of the moving object, thereby realizing efficient, smooth and obstacle-avoiding safe path prediction for the next position point of the moving object, thereby improving the real-time and reliability of path planning.

[0088] Each step is explained in detail below.

[0089] S101: Obtain obstacle information and position information of a moving object; the position information includes target position coordinate information, current position coordinate information, and historical position coordinate information;

[0090] Specifically, the server acquires input information required for path prediction, including obstacle information, target position coordinate information of the moving object, current position coordinate information, and historical position coordinate information. The obstacle information is used to represent the spatial distribution of obstacles existing in the current environment; the target position coordinate information is the coordinate of the target position point; the current position information is the coordinate of the current position point; and the historical position coordinate information includes a plurality of historical moving position points.

[0091] S102: generating first state feature information of the current position according to the target position coordinate information, the current position coordinate information, the historical position coordinate information, and the obstacle information;

[0092] Specifically, the server constructs state feature information of the current position based on the above input information (including obstacle information, target position coordinate information of the moving object, current position coordinate information, and historical position coordinate information). The feature information reflects the relative relationship between the current position and the environment and the target, and is used to depict the state performance of the moving object at the current position.

[0093] Figure 12 is a path planning schematic diagram provided by an embodiment of the present application, as shown in Figure 12 , the vehicle starts moving from P Start , passes through intermediate position points P T-1 , P T , P T+1 , and finally drives to the end point P Goal . The black boxes in the figure represent obstacles, and the vehicle maintains a safe distance from the obstacles during driving. The path planning method proposed in the present application is based on the current position point P T and the historical position points P T-1 , …, P T-n+1 to predict the next position P T+1 , where n is the number of historical position points. The line P T P T-1 connecting the current position point P T-1 and the position point P T at time T-1 has an angle β with the line P T P T+1 , and the line P T-1 P T connecting the current position point P T and the position point P Goal has an angle ɑ with the line P T P T+1 .

[0094] Figure 2 is a flowchart of a path prediction method based on a time sequence convolution network provided by an embodiment of the present application, and the historical position coordinate information includes the last moving position coordinate information of the current position of the moving object; as shown in Figure 2 , S102 includes:

[0095] S201: Calculating a first average Euclidean distance from the current position to the obstacle based on the current position coordinate information and the obstacle information;

[0096] Specifically, when constructing the state feature information for the current position, not only the target position coordinates, the current position coordinates, and obstacle information are considered, but also the coordinates of the previous position before the current position are incorporated. This historical point is used to determine the current direction of movement, thereby enhancing the modeling capability of path directionality.

[0097] Based on the spatial distribution of the current location and obstacles, the server calculates the Euclidean distances from the current location to multiple obstacles and takes the average of these distances as the average Euclidean distance from the current location to the obstacles. This feature reflects the obstacle avoidance risk level at the current location.

[0098] In one embodiment, the moving object is simplified into a mass point, and the path planning problem is transformed into finding a collision-free optimal path for a point from a starting point to a target point in the motion space. The target position point is determined, and the moving object moves from the starting position point to the target position point. Figure 12 As shown, the current position point P of the moving object T , its absolute position coordinates , each time moving a fixed distance d towards the target point R , and As the center of the circle, with d R Draw a circle O with a radius of 0. The obstacles within the circle are effective obstacles. T+1 is located at the boundary of the circle. As the center of the circle, with d R Circle P as the radius T+1 The effective obstacles within the range will affect the next moving position point P T+1 Choice, P T+1 Collisions with all valid obstacles should be avoided as much as possible. Assume that the number of valid obstacles is M, P T The Euclidean distance to obstacle i is d oui , P T Average distance to all valid obstacles Expressed as:

[0099]

[0100] S202: Calculating a first Euclidean distance from the current position to the target position based on the current position coordinate information and the target position coordinate information;

[0101] Specifically, the server calculates the Euclidean distance from the current position to the target position according to the coordinate difference between the current position point and the target position point, for quantifying the closeness to the target position.

[0102] In an embodiment, the Euclidean distance from the current position to the target position is calculated as:

[0103]

[0104] S203: Determine a first included angle between the heading and the line connecting the target point according to the current position coordinate information, the last moving position coordinate information and the target position coordinate information.

[0105] Specifically, the server determines the current moving direction according to the line direction P T-1 P T connecting the current position and the last moving position point, and then compares the direction with the line direction P T P Goal connecting the current position and the target position, calculates the included angle therebetween, and obtains the deviation between the current moving direction and the target direction. The included angle can reflect whether the moving object is currently moving towards the target direction, and has guiding significance for path smoothness and target approaching.

[0106] S204: Construct the first state feature information by using the current position coordinate information, the first average Euclidean distance, the first Euclidean distance and the first included angle.

[0107] Specifically, the server combines the current position coordinate information, the average Euclidean distance from the current position to the obstacle, the Euclidean distance from the current position to the target point and the included angle between the current moving direction and the target direction as the state feature information of the current position, and uses the state feature information as the input of the subsequent path prediction model.

[0108] In an embodiment, the state feature information of the current position constructed is represented as:

[0109]

[0110] Through the above processing mode, the application can more accurately depict the moving environment and target guiding features of the current position, provide more discriminative input data for the path prediction model, and thus improve the accuracy and stability of the prediction effect.

[0111] S103: Generate second state feature information of a plurality of historical positions according to the target position coordinate information, the historical position coordinate information and the obstacle information.

[0112] ​​Specifically, the server constructs state feature information of a plurality of historical position points based on the target position, the historical trajectory, and the obstacle information, to reflect the behavior pattern of the moving object in the past period of time and the relationship between the moving object and the target and the environment.

[0113] Figure 3 is a flowchart of a path prediction method based on a time sequence convolution network provided by an embodiment of the present application, as shown in Figure 3 S103 includes:

[0114] S301: calculating a second average Euclidean distance from each historical position to an obstacle according to the plurality of historical position coordinate information and the obstacle information;

[0115] Specifically, in order to enhance the time sequence understanding ability of the path prediction model for the historical trajectory, the server constructs the state feature information of each historical position point, similar to the construction of the state feature information of the current position, so that the model can capture the trend of the change of the moving pattern in the time dimension, and improve the stability and continuity of the path prediction.

[0116] The server calculates the Euclidean distance between each historical position point and each obstacle based on the coordinate information of the plurality of historical position points and the spatial distribution information of the obstacles in the current environment, and performs average processing on the Euclidean distances, to obtain the average Euclidean distance from each historical position point to the obstacle. This index can reflect the spaciousness of the obstacle avoidance of the position at this moment, which is helpful for the model to understand the obstacle avoidance behavior in the trajectory evolution.

[0117] S302: calculating a second Euclidean distance from each historical position to the target position according to the plurality of historical position coordinate information and the target position coordinate information;

[0118] Specifically, the server calculates the Euclidean distance from each historical point to the target point according to the coordinate difference between each historical position point and the target position point, to describe the proximity of the historical point to the target position. This information can be used to guide the model to identify the convergence trend of the trajectory and distinguish between effective paths and deviated paths.

[0119] S303: determining a second included angle between the heading of each historical position and the line connecting the target point according to the plurality of historical position coordinate information and the target position coordinate information;

[0120] Specifically, to depict the evolution process of the trajectory direction, for each historical position point, the server extracts its own and its previous historical point, constructs a moving direction vector, and then calculates the included angle between the vector and the direction vector of the line connecting the historical point to the target position point, to obtain the included angle between the heading of each historical position point and the line connecting the target. The included angle value can be used to measure whether the path is moving towards the target at this moment, and reflect the performance of the trajectory in the target direction.

[0121] S304: Construct a plurality of second state feature information using the historical position coordinate information, the second average Euclidean distance, the second Euclidean distance, and the second included angle.

[0122] Specifically, the server combines the historical position coordinate information with the corresponding historical position to obstacle average Euclidean distance, historical position to target point Euclidean distance, and the included angle between the heading and the target line as the state feature information of each historical position point. The state features of all historical position points are arranged in time sequence to form the time sequence input feature sequence required by the path prediction model.

[0123] In the above manner, the historical trajectory is fully exploited for its dynamic characteristics in obstacle avoidance behavior, target proximity, and direction control, which helps to improve the modeling capability of the model for time sequence behavior rules, thereby achieving higher precision and more robust path prediction results.

[0124] S104: Construct a feature information sequence combining the first state feature information and the second state feature information;

[0125] Specifically, the server combines the state feature information of the current position with the state feature information of a plurality of historical positions in time sequence to construct a feature information sequence. This sequence retains the time sequence dependency and is the main input data of the path prediction model.

[0126] In an embodiment, the feature information sequence L T consists of the information of the current position point P T-1 , …, P T-n+1 N points, and the feature information sequence L T is represented as follows:

[0127]

[0128] Combining formula (3) and formula (4), the following information matrix is obtained:

[0129]

[0130] L T is input as data into the path prediction model PathPreTCN based on TCN, and PathPreTCN predicts the next position point P T+1 . The path prediction method based on historical position points is a heuristic path prediction method.

[0131] S105: Input the feature information sequence into the pre-generated path prediction model to obtain the next moving position coordinate information of the moving object.

[0132] Specifically, the server inputs the feature information sequence into a pre-generated path prediction model, and outputs a prediction result, i.e., the moving position coordinate information of the moving object at the next moment. The prediction result can be used to guide the navigation action of the object at the next step, thereby forming a complete and continuous path planning process.

[0133] In an embodiment, the overall flow of the heuristic path prediction method based on the TCN network is as shown in Figure 13

[0134] Through the above steps, the application can realize effective prediction of future path points based on historical behavior and current position state, and improve path continuity, obstacle avoidance capability and target convergence efficiency.

[0135] Figure 4 is a flowchart of the path prediction method based on the temporal convolutional network provided by an embodiment of the application, as shown in Figure 4 The step of pre-generating the path prediction model includes:

[0136] S401: An initial model is constructed using a temporal convolutional network structure, and the initial model includes a fully connected layer, a temporal convolutional network module, a feature shaping module and a multilayer perceptron module;

[0137] Specifically, to realize efficient prediction of the path of the moving object, a path prediction model is pre-constructed and trained, and the model is based on a temporal convolutional network (TCN) structure and has strong time series modeling capability and prediction stability.

[0138] The server constructs an initial model, and the model as a whole adopts a TCN structure and includes multiple functional modules: the input feature information is first linearly transformed and dimensionally unified by a fully connected layer; then it enters a temporal convolutional network module, which models the time dependence of the historical trajectory through a dilated causal convolution structure while preserving causality; then, the output feature is dimensionally adjusted and compressed by a feature shaping module for matching the input requirements of the subsequent structure; finally, the shaped feature is sent to a multilayer perceptron (MLP) module for realizing nonlinear fitting and outputting the prediction position coordinate at the next moment.

[0139] S402: The initial model is trained using pre-generated training samples to predict the next moving position coordinate information of the moving object;

[0140] ​Specifically, the server trains the initial model using pre-built training samples. During the training phase, the model takes a feature information sequence as input and generates predictions through forward propagation. The training samples are pre-built feature information training sequences based on the acquired training data using the same method used to construct the feature information sequence.

[0141] S403: Calculate the prediction error using a pre-built composite loss model, and optimize the parameters of the trained initial model through back propagation until the model converges, thereby obtaining the path prediction model.

[0142] Specifically, to improve model prediction performance, a composite loss model is introduced as the objective function during training. This performs a comprehensive, multi-dimensional evaluation of prediction errors, and backpropagation is performed based on these errors to iteratively optimize model parameters. Training continues until the model's performance on the validation set reaches a preset standard or the error converges. This ultimately results in the heuristic path prediction model, PathPreTCN, for online prediction.

[0143] In one embodiment, the architecture of the heuristic path prediction model PathPreTCN is as follows: Figure 14 As shown, the fully connected layer Linear Layers mainly implements information sequence The input projection matches the weight of each piece of information. The Temporal Convolutional Network (TCN) module adopts a dilated causal convolution architecture, achieving exponential expansion of the receptive field through multiple layers of cascaded dilated convolution kernels. This effectively captures long-range dependencies while maintaining strict temporal causality. This design ensures that when processing time series data, the model relies solely on historical information for future predictions. Combined with residual connections and batch normalization techniques, it further enhances gradient propagation efficiency and prediction accuracy. The feature reshaping module adjusts the information features output by the TCN module to meet the input requirements of the subsequent multi-layer perceptron (MLP) module. The MLP module integrates the predicted information output by the TCN according to task requirements and ultimately outputs the information for the next location point.

[0144] Through the above modeling and training process, the generated path prediction model can fully exploit the temporal correlation characteristics between historical trajectories and goal guidance, realize accurate prediction of the next position point of the moving object, and provide key support for path planning.

[0145] Figure 5 This is a flow chart of a path prediction method based on a temporal convolutional network provided by an embodiment of the present application. Figure 5 As shown, the steps of pre-building the compound loss model include:

[0146] S501: Generate a first superposition control item according to a third Euclidean distance from the predicted position to the target position and a first Euclidean distance from the current position to the target position;

[0147] Specifically, in order to achieve a multi-dimensional comprehensive evaluation of the path prediction model training effect, a composite loss model was constructed to measure the comprehensive performance of the predicted path in terms of target approach, direction rationality and obstacle avoidance safety, thereby guiding the model training to be closer to actual navigation needs.

[0148] The server compares the Euclidean distance between the predicted position point and the target position point with the Euclidean distance from the current position to the target position point to obtain a first superposition control item, which is used to measure whether the predicted point is closer to the target point than the current position, reflecting the target convergence of the path.

[0149] In one embodiment, the server constructs a composite loss model Loss total Compound loss model Loss total Factors affected include forecast points Weighted distance to all valid obstacles , the angle between the predicted heading and the line connecting the target point and forecast points Euclidean distance to the target point . Forecast point Information collection .

[0150] Prediction Point and the target point P Goal Euclidean distance With the current position point P T Euclidean distance to the target point The first superposition control term of the composite loss model is composed of the ratio of It is expressed as follows:

[0151]

[0152] The smaller the loss, the better the prediction point. The closer to the target point P Goal , and the purpose of path planning is to approach the target point by the shortest path.

[0153] S502: generating a second superposition control item according to a third angle between a line connecting the current position and the predicted position and a line connecting the current position and the target position;

[0154] Specifically, the server extracts the line direction from the current position to the predicted position point and the line direction from the current position to the target position point, calculates the included angle between the two, obtains a second superimposed control item, which is used to measure whether the movement direction of the predicted point is towards the target direction, helps to keep the correct heading of the path, and reduces unnecessary deviation.

[0155] In an embodiment, the current position point P T and the predicted point P are connected by a line , and the line from the current position point P T to the target point P Goal is P T P Goal , and the included angle between P T P Goal is , which is a second superimposed control item of the composite loss model , and is represented as follows:

[0156] (7)

[0157] The smaller the value is, the smoother the path is, and a smoother and straighter driving path is expected in the path planning process.

[0158] S503: generating a third superimposed control item according to the third average Euclidean distance from the predicted position to the obstacle;

[0159] Specifically, the server calculates the Euclidean distance between the predicted position point and multiple obstacles in the environment, and obtains a weighted average value, to obtain a third superimposed control item, which is used to measure the spatial distance between the predicted point and the obstacle, and ensure the obstacle avoidance safety of the path.

[0160] In an embodiment, the weighted distance from the predicted point P to all effective obstacles is , and there are M effective obstacles within a range with the predicted point P as the center and d R as the radius, and the position of the i-th effective obstacle is , which is a third superimposed control item of the composite loss model , and is represented as follows:

[0161] (8)

[0162] When the moving object reaches the predicted point position, it is expected to have a certain distance from all obstacles to avoid collision. The smaller the value is, the smaller the probability of collision is.

[0163] S504: constructing the composite loss model according to the pre-constructed initial loss model, the first superimposed control item, the second superimposed control item and the third superimposed control item.

[0164] Specifically, the server combines the preset initial loss model with the above three superimposed control items to form a final composite loss model. The composite loss model serves as an optimization target in the training process to guide the path prediction model to continuously adjust parameters, so that the prediction result simultaneously obtains better performance in terms of approaching the target, avoiding obstacles and maintaining reasonable direction.

[0165] In an embodiment, the initial loss model used is a Huber loss model Loss Huber Huber :

[0166] (9)

[0167] Based on the analysis of the above loss functions, the composite loss model Loss TCN of the heuristic path prediction model based on the TCN network is constructed as follows: total

[0168] (10)

[0169] wherein k1, k2 and k3 represent the weight values of the loss function superimposed control items , and respectively, and satisfy the following constraint:

[0170] k1+k2+k3=1 (11)

[0171] The composite loss model constructed by the above method can comprehensively constrain and optimize the path prediction quality from multiple dimensions, and has stronger engineering practicability and prediction robustness compared with the traditional single loss function.

[0172] ​The application provides a path prediction method based on a time sequence convolution network, which comprises the following steps: obtaining obstacle information and position information of a moving object; the position information comprises target position coordinate information, current position coordinate information and historical position coordinate information; generating first state feature information of a current position according to the target position coordinate information, the current position coordinate information, the historical position coordinate information and the obstacle information; generating second state feature information of a plurality of historical positions according to the target position coordinate information, the historical position coordinate information and the obstacle information; combining the first state feature information and the second state feature information to construct a feature information sequence; and inputting the feature information sequence into a pre-generated path prediction model to obtain next moving position coordinate information of the moving object, thereby realizing efficient, smooth and obstacle-avoiding safe path prediction of a next position point of the moving object and improving real-time performance and reliability of path planning.

[0173] In the method, the obstacle information and the position information of the moving object are obtained, the position information comprises the target position coordinate information, the current position coordinate information and the historical position coordinate information, comprehensive acquisition of environment perception and motion state data required for path prediction is realized, the first state feature information of the current position is generated according to the target position coordinate information, the current position coordinate information, the historical position coordinate information and the obstacle information, quantitative expression of comprehensive motion state and environmental risk of the current position is realized, the second state feature information of the plurality of historical positions is generated according to the target position coordinate information, the historical position coordinate information and the obstacle information, state modeling of each point in the historical trajectory in terms of target direction and obstacle-avoiding safety is realized, the feature information sequence is constructed by combining the first state feature information and the second state feature information, time sequence motion state information is structured into an input sequence that can be used for model prediction, the next moving position coordinate information of the moving object is obtained by inputting the feature information sequence into the pre-generated path prediction model, and intelligent prediction of the next moving position of the moving object is realized.

[0174] Based on the same inventive concept, the application also provides a path prediction device based on a time sequence convolution network, which can be used to realize the method described in the above embodiments, as described in the following embodiments. Since the path prediction device based on the time sequence convolution network solves problems in a similar principle to the path prediction method based on the time sequence convolution network, the implementation of the path prediction device based on the time sequence convolution network can be referred to the implementation of the method based on the software performance benchmark, and the repeated parts will not be described herein. The term "unit" or "module" used below can be a combination of software and / or hardware that realizes a predetermined function. Although the system described in the following embodiments is preferably realized in software, hardware or a combination of software and hardware is also possible and is conceived.

[0175] Figure 6 is a structural schematic diagram of a path prediction device based on a time sequence convolution network provided by an embodiment of the present application, as shown in the figure, the path prediction device based on the time sequence convolution network provided by the present application comprises: Figure 6

[0176] an information acquisition unit 601, configured to acquire obstacle information and position information of a moving object; the position information comprises target position coordinate information, current position coordinate information and historical position coordinate information;

[0177] a current state feature information generation unit 602, configured to generate first state feature information of a current position according to the target position coordinate information, the current position coordinate information, the historical position coordinate information and the obstacle information;

[0178] a historical state feature information generation unit 603, configured to generate second state feature information of a plurality of historical positions according to the target position coordinate information, the historical position coordinate information and the obstacle information;

[0179] a feature information sequence construction unit 604, configured to construct a feature information sequence in combination with the first state feature information and the second state feature information;

[0180] a path prediction unit 605, configured to input the feature information sequence into a path prediction model generated in advance to obtain next moving position coordinate information of the moving object.

[0181] Figure 7 is a structural schematic diagram of a path prediction device based on a time sequence convolution network provided by an embodiment of the present application, on the basis of the embodiment, further, as shown in the figure, the current state feature information generation unit 602 comprises: Figure 6 Figure 7

[0182] an obstacle distance calculation module 701, configured to calculate a first average Euclidean distance from the current position to an obstacle according to the current position coordinate information and the obstacle information;

[0183] a target position distance calculation module 702, configured to calculate a first Euclidean distance from the current position to a target position according to the current position coordinate information and the target position coordinate information;

[0184] an included angle calculation module 703, configured to determine a first included angle of a heading and a line connecting the target point according to the current position coordinate information, the last moving position coordinate information and the target position coordinate information;

[0185] ​​​The current state feature information generation module 704 is configured to construct the first state feature information by using the current position coordinate information, the first average Euclidean distance, the first Euclidean distance, and the first included angle.

[0186] Figure 8 is a structural schematic diagram of a path prediction device based on a time sequence convolution network provided in an embodiment of the present application, in Figure 6 On the basis of the embodiments, further, as Figure 8 indicated, the historical state feature information generation unit 603 includes:

[0187] The historical obstacle distance calculation module 801 is configured to calculate a second average Euclidean distance from each historical position to an obstacle according to the plurality of historical position coordinate information and the obstacle information.

[0188] The historical target position distance calculation module 802 is configured to calculate a second Euclidean distance from each historical position to a target position according to the plurality of historical position coordinate information and the target position coordinate information.

[0189] The historical included angle calculation module 803 is configured to determine a second included angle of a heading of each historical position and a line connecting the target point according to the plurality of historical position coordinate information and the target position coordinate information.

[0190] The historical state feature information generation module 804 is configured to construct a plurality of second state feature information by using the historical position coordinate information, the second average Euclidean distance, the second Euclidean distance, and the second included angle.

[0191] Figure 9 is a structural schematic diagram of a path prediction device based on a time sequence convolution network provided in an embodiment of the present application, in Figure 6 On the basis of the embodiments, further, as Figure 9 indicated, the path prediction device based on a time sequence convolution network further includes:

[0192] The model construction unit 901 is configured to construct an initial model by using a time sequence convolution network structure, and the initial model includes a full connection layer, a time sequence convolution network module, a feature shaping module, and a multi-layer perception module.

[0193] The model training unit 902 is configured to train the initial model by using a pre-generated training sample, so as to predict a next moving position coordinate information of a moving object.

[0194] The model optimization unit 903 is configured to calculate a prediction error by using a pre-constructed compound loss model, and optimize parameters of the trained initial model by back propagation until the model converges, so as to obtain the path prediction model.

[0195] Figure 10 is a structural schematic diagram of a path prediction device based on a timing convolution network provided by an embodiment of the present application, in which Figure 9 on the basis of the embodiment, further, as shown in Figure 10 the path prediction device based on the timing convolution network further comprises:

[0196] a first superposition control item generation unit 1001 configured to generate a first superposition control item according to a third Euclidean distance from a predicted position to a target position and a first Euclidean distance from a current position to the target position;

[0197] a second superposition control item generation unit 1002 configured to generate a second superposition control item according to a third included angle between a line connecting the current position and the predicted position and a line connecting the current position and the target position;

[0198] a third superposition control item generation unit 1003 configured to generate a third superposition control item according to a third average Euclidean distance from the predicted position to an obstacle;

[0199] a composite loss model construction unit 1004 configured to construct the composite loss model according to a pre-constructed initial loss model, the first superposition control item, the second superposition control item, and the third superposition control item.

[0200] The present application provides a path prediction method and device based on a timing convolution network, which obtains obstacle information and position information of a moving object; the position information includes target position coordinate information, current position coordinate information, and historical position coordinate information; generates first state feature information of the current position according to the target position coordinate information, the current position coordinate information, the historical position coordinate information, and the obstacle information; generates second state feature information of a plurality of historical positions according to the target position coordinate information, the historical position coordinate information, and the obstacle information; constructs a feature information sequence in combination with the first state feature information and the second state feature information; inputs the feature information sequence into a pre-generated path prediction model to obtain next movement position coordinate information of the moving object, thereby realizing efficient, smooth, and obstacle-avoiding safe path prediction of the next position point of the moving object and improving real-time performance and reliability of path planning.

[0201] Wherein, by acquiring obstacle information and position information of the moving object; the position information includes target position coordinate information, current position coordinate information and historical position coordinate information, realizing comprehensive acquisition of environment perception and motion state data required for path prediction; by generating first state feature information of the current position according to the target position coordinate information, the current position coordinate information, the historical position coordinate information and the obstacle information, realizing quantitative expression of comprehensive motion state and environmental risk of the current position; by generating second state feature information of a plurality of historical positions according to the target position coordinate information, the historical position coordinate information and the obstacle information, realizing state modeling of each point in the historical trajectory in terms of target orientation and obstacle avoidance safety; by combining the first state feature information and the second state feature information to construct a feature information sequence, realizing structuring of time-series motion state information into an input sequence that can be used for model prediction; by inputting the feature information sequence into a pre-generated path prediction model, obtaining next moving position coordinate information of the moving object, realizing intelligent prediction of the next moving position of the moving object.

[0202] From the hardware level, in order to solve the problems in the prior art, an embodiment of an electronic device for implementing all or part of the contents of the path prediction method based on the time-series convolutional network is provided, and the electronic device specifically includes the following contents:

[0203] A processor (Processor), a memory (Memory), a communications interface (Communications Interface) and a bus; wherein the processor, the memory and the communications interface complete mutual communication through the bus; the communications interface is used to realize information transmission between the path prediction device based on the time-series convolutional network and related devices such as core business systems, user terminals and related databases; the logic controller can be a desktop computer, a tablet computer and a mobile terminal, and the embodiment is not limited thereto. In the embodiment, the logic controller can be implemented with reference to the embodiment of the path prediction method based on the time-series convolutional network and the embodiment of the path prediction device based on the time-series convolutional network, the contents of which are incorporated herein, and repeated descriptions are omitted.

[0204] It can be understood that the user terminal can include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, etc. The smart wearable device can include smart glasses, a smart watch, a smart bracelet, etc.

[0205] In practical applications, part of the path prediction method based on the time sequence convolution network can be executed on the electronic device as described above, or all operations can be completed in the client device. Specifically, the processing capacity of the client device and the limitations of the user's use scenario can be selected. The present application does not limit this. If all operations are completed in the client device, the client device can also include a processor.

[0206] The above-mentioned client device can have a communication module (i.e. a communication unit) that can be in communication connection with a remote server to realize data transmission with the server. The server can include a server on the task scheduling center side, and other implementation scenarios can also include a server of an intermediate platform, such as a server of a third-party server platform that is in communication link with the task scheduling center server. The server can include a single computer device, or a server cluster composed of multiple servers, or a server structure of a distributed device.

[0207] Figure 11 The schematic block diagram of the system structure of the electronic device 9600 of the embodiment of the present application is shown in FIG. 9. As shown in the figure, the electronic device 9600 can include a central processor 9100 and a memory 9140; the memory 9140 is coupled to the central processor 9100. It is worth noting that the structure shown in the figure is exemplary; other types of structures can also be used to supplement or replace the structure to realize telecommunication functions or other functions. Figure 11 Figure 11 The structure shown in the figure is exemplary; other types of structures can also be used to supplement or replace the structure to realize telecommunication functions or other functions.

[0208] In an embodiment, the path prediction method based on the time sequence convolution network can be integrated into the central processor 9100. The central processor 9100 can be configured to control as follows:

[0209] S101: Obtain obstacle information and position information of a moving object; the position information includes target position coordinate information, current position coordinate information and historical position coordinate information;

[0210] S102: Generate first state feature information of the current position according to the target position coordinate information, the current position coordinate information, the historical position coordinate information and the obstacle information;

[0211] S103: Generate second state feature information of a plurality of historical positions according to the target position coordinate information, the historical position coordinate information and the obstacle information;

[0212] S104: Construct a feature information sequence in combination with the first state feature information and the second state feature information;

[0213] ​S105: Input the feature information sequence into a pre-generated path prediction model to obtain the next moving position coordinate information of the moving object.

[0214] From the above description, it can be seen that the path prediction method and device based on the temporal convolutional network provided in this application realize efficient, smooth and obstacle-avoiding path prediction of the next position point of the moving object, thereby improving the real-time and reliability of path planning.

[0215] In another embodiment, the path prediction device based on the temporal convolutional network can be configured separately from the central processing unit 9100. For example, the path prediction device based on the temporal convolutional network of the data composite transmission device can be configured as a chip connected to the central processing unit 9100, and the function of the path prediction method based on the temporal convolutional network can be realized through the control of the central processing unit.

[0216] like Figure 11 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily have to include Figure 11 In addition, the electronic device 9600 may also include all components shown in Figure 11 For components not shown, reference may be made to the prior art.

[0217] like Figure 11 As shown, the central processing unit 9100 is sometimes also referred to as a controller or operation control, and may include a microprocessor or other processor device and / or logic device. The central processing unit 9100 receives input and controls the operation of various components of the electronic device 9600.

[0218] Memory 9140 can be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It can store the aforementioned failure-related information and also store programs that execute the relevant information. The CPU 9100 can execute the programs stored in memory 9140 to implement information storage or processing.

[0219] The input unit 9120 provides input to the central processing unit 9100. The input unit 9120 may be, for example, a keypad or touch input device. The power supply 9170 is used to provide power to the electronic device 9600. The display 9160 is used to display objects such as images and text. The display may be, for example, an LCD display, but is not limited thereto.

[0220] The memory 9140 can be a solid state memory such as a read only memory (ROM), a random access memory (RAM), a SIM card, or the like. It can also be a memory that retains information even when power is off, can be selectively erased, and is provided with more data, an example of which is sometimes referred to as an EPROM or the like. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 can include an application / function storage 9142 for storing application programs and function programs or for storing a flow for executing an operation of the electronic device 9600 by the central processing unit 9100.

[0221] The memory 9140 can also include a data storage 9143 for storing data such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. A driver storage 9144 of the memory 9140 can include various drivers of the electronic device for a communication function and / or for performing other functions of the electronic device such as a messaging application, a phonebook application, and the like.

[0222] The communication module 9110 is a transmitter / receiver 9110 that transmits and receives signals via an antenna 9111. The communication module (transmitter / receiver) 9110 is coupled to the central processing unit 9100 to provide input signals and receive output signals, which can be the same as in the case of a conventional mobile communication terminal.

[0223] Based on different communication technologies, a plurality of communication modules 9110 such as a cellular network module, a Bluetooth module, and / or a wireless local area network module, and the like can be provided in the same electronic device. The communication module (transmitter / receiver) 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby implementing a conventional telecommunication function. The audio processor 9130 can include any suitable buffer, decoder, amplifier, and the like. In addition, the audio processor 9130 is also coupled to the central processing unit 9100, thereby enabling recording on the local device through the microphone 9132 and enabling playing of a sound stored on the local device through the speaker 9131.

[0224] The embodiment of the present application further provides a computer readable storage medium capable of realizing all steps of the path prediction method based on the time series convolution network with the execution subject being the server or the client in the above embodiment, and the computer program is stored on the computer readable storage medium, and when the computer program is executed by a processor, all steps of the path prediction method based on the time series convolution network with the execution subject being the server or the client in the above embodiment are realized, for example, the following steps are realized when the processor executes the computer program:

[0225] S101: obtain obstacle information and position information of a moving object; the position information comprises target position coordinate information, current position coordinate information and historical position coordinate information;

[0226] S102: generate first state feature information of a current position according to the target position coordinate information, the current position coordinate information, the historical position coordinate information and the obstacle information;

[0227] S103: generate second state feature information of a plurality of historical positions according to the target position coordinate information, the historical position coordinate information and the obstacle information;

[0228] S104: construct a feature information sequence in combination with the first state feature information and the second state feature information;

[0229] S105: input the feature information sequence into a pre-generated path prediction model to obtain next movement position coordinate information of the moving object.

[0230] From the above description, it can be known that the path prediction method and device based on the time series convolution network provided by the present application realize efficient, smooth and obstacle-avoiding safe path prediction of a next position point of a moving object, and improve real-time performance and reliability of path planning.

[0231] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, device or computer program product. Therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can adopt a computer program product in the form of being implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0232] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (apparatus), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0233] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0234] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0235] Specific embodiments are used in this application to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A path prediction method based on a temporal convolutional network, characterized in that: include: Obtain obstacle information and position information of moving objects; The location information includes target location coordinate information, current location coordinate information and historical location coordinate information; generating first state characteristic information of the current position according to the target position coordinate information, the current position coordinate information, the historical position coordinate information, and the obstacle information; generating second state feature information of a plurality of historical positions according to the target position coordinate information, the historical position coordinate information, and the obstacle information; constructing a characteristic information sequence by combining the first state characteristic information and the second state characteristic information; The feature information sequence is input into a pre-generated path prediction model to obtain the next moving position coordinate information of the moving object.

2. The path prediction method based on a temporal convolutional network according to claim 1, characterized in that: The historical position coordinate information includes the previous moving position coordinate information of the current position of the moving object; the first state feature information of the current position is generated according to the target position coordinate information, the current position coordinate information, the historical position coordinate information and the obstacle information, including: Calculating a first average Euclidean distance from the current position to the obstacle based on the current position coordinate information and the obstacle information; Calculate a first Euclidean distance from the current position to the target position based on the current position coordinate information and the target position coordinate information; Determine a first angle between the heading and the line connecting the target point according to the current position coordinate information, the last moving position coordinate information and the target position coordinate information; The first state feature information is constructed using the current position coordinate information, the first average Euclidean distance, the first Euclidean distance, and the first angle.

3. The path prediction method based on a temporal convolutional network according to claim 1, characterized in that: The generating of the second state characteristic information of the plurality of historical positions according to the target position coordinate information, the historical position coordinate information and the obstacle information includes: Calculating a second average Euclidean distance from each historical position to the obstacle based on the plurality of historical position coordinate information and the obstacle information; Calculate the second Euclidean distance from each historical position to the target position based on the plurality of historical position coordinate information and the target position coordinate information; Determine a second angle between the course of each historical position and the line connecting the target point based on the plurality of historical position coordinate information and the target position coordinate information; A plurality of second state feature information is constructed using the historical position coordinate information, the second average Euclidean distance, the second Euclidean distance, and the second angle.

4. The path prediction method based on a temporal convolutional network according to claim 1, characterized in that: The step of pre-generating the path prediction model includes: An initial model is constructed using a temporal convolutional network structure, wherein the initial model includes a fully connected layer, a temporal convolutional network module, a feature shaping module, and a multi-layer perceptron module; The initial model is trained using pre-generated training samples to predict the next moving position coordinate information of the moving object; The prediction error is calculated using a pre-built composite loss model, and the parameters of the trained initial model are optimized through back propagation until the model converges, thereby obtaining the path prediction model.

5. The path prediction method based on a temporal convolutional network according to claim 4, characterized in that: The steps of pre-building the compound loss model include: generating a first superposition control item according to a third Euclidean distance from the predicted position to the target position and a first Euclidean distance from the current position to the target position; generating a second superposition control item according to a third angle between a line connecting the current position and the predicted position and a line connecting the current position and the target position; generating a third superposition control item according to a third average Euclidean distance from the predicted position to the obstacle; The composite loss model is constructed according to the pre-constructed initial loss model, the first superposition control item, the second superposition control item and the third superposition control item.

6. A path prediction device based on a temporal convolutional network, characterized in that: include: An information acquisition unit, used to acquire obstacle information and position information of moving objects; The location information includes target location coordinate information, current location coordinate information and historical location coordinate information; a current state characteristic information generating unit, configured to generate first state characteristic information of the current position based on the target position coordinate information, the current position coordinate information, the historical position coordinate information, and the obstacle information; a historical state characteristic information generating unit, configured to generate second state characteristic information of a plurality of historical positions according to the target position coordinate information, the historical position coordinate information, and the obstacle information; a characteristic information sequence construction unit, configured to construct a characteristic information sequence by combining the first state characteristic information and the second state characteristic information; The path prediction unit is used to input the feature information sequence into a pre-generated path prediction model to obtain the next moving position coordinate information of the moving object.

7. The path prediction device based on a temporal convolutional network according to claim 6, characterized in that: The historical position coordinate information includes the previous moving position coordinate information of the current position of the moving object; The current state characteristic information generating unit includes: an obstacle distance calculation module, configured to calculate a first average Euclidean distance from the current position to the obstacle based on the current position coordinate information and the obstacle information; a target position distance calculation module, configured to calculate a first Euclidean distance from the current position to the target position based on the current position coordinate information and the target position coordinate information; An angle calculation module, used to determine a first angle between the heading and the line connecting the target point according to the current position coordinate information, the last moving position coordinate information and the target position coordinate information; The current state feature information generating module is configured to construct the first state feature information by using the current position coordinate information, the first average Euclidean distance, the first Euclidean distance, and the first angle.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images