WO2023103142A1

WO2023103142A1 - Obstacle trajectory prediction method and system, and computer-readable storage medium

Info

Publication number: WO2023103142A1
Application number: PCT/CN2022/071090
Authority: WO
Inventors: 黄超; 黎罗河
Original assignee: 上海仙途智能科技有限公司
Priority date: 2021-12-07
Filing date: 2022-01-10
Publication date: 2023-06-15
Also published as: CN115230688A; CN115230688B

Abstract

Provided in the present application are an obstacle trajectory prediction method and system, and a computer-readable storage medium. The obstacle trajectory prediction method comprises: acquiring information of an obstacle and map information of a predicted motion of the obstacle; determining a plurality of candidate destinations of the obstacle according to the map information; extracting features of the plurality of candidate destinations according to the plurality of candidate destinations, extracting map features according to the map information, and extracting obstacle features according to the information of the obstacle; determining a plurality of predetermined destinations from the plurality of candidate destinations according to these features, wherein compared with the remaining candidate destinations from among the plurality of candidate destinations, the plurality of predetermined destinations have a higher possibility of becoming a target destination of a predicted motion trajectory; determining target destinations of the obstacle according to the plurality of predetermined destinations, wherein the target destinations comprise the destination of a historical trajectory and a future arrival destination; and according to the target destinations, determining a predicted motion trajectory of the obstacle from the destination of the historical trajectory to the future arrival destination.

Description

Method, system and computer-readable storage medium for obstacle trajectory prediction

technical field

The present application relates to the technical field of automatic driving, and in particular to a method, system and computer-readable storage medium for obstacle trajectory prediction.

Background technique

With the maturity of self-driving technology, self-driving vehicles can be widely used. In order to enable self-driving vehicles to drive in complex scenes such as urban roads, closed parks, and expressways, self-driving vehicles need to predict the trajectory of surrounding vehicles, pedestrians, bicycles and other obstacles in order to predict risks and avoid accidents.

In the method of the related art, the map information related to the obstacle and the historical track information of the obstacle are preprocessed, and then input into the neural network, so as to determine the target track of the obstacle. However, the method of the related art is a black-box model, and its internal working mechanism is difficult to be understood by developers, which is not conducive to the later optimization of the method, and it is not convenient to grasp the determination process of the estimated motion trajectory in time.

Contents of the invention

The present application provides a method, system and computer-readable storage medium for obstacle trajectory prediction, so as to facilitate the understanding of developers.

The present application provides a method for predicting the trajectory of an obstacle, including: obtaining information about the obstacle and map information of the estimated movement of the obstacle; determining multiple candidate end points of the obstacle according to the map information; The candidate terminal extracts a plurality of features of the candidate terminal, extracts the map feature according to the map information, and extracts the obstacle feature according to the information of the obstacle; according to the feature of the candidate terminal, the map feature and the obstacle characteristics, determining a plurality of predetermined end points from among the plurality of candidate end points, and the plurality of predetermined end points become the target end points of the estimated motion trajectory compared with the remaining candidate end points among the plurality of candidate end points The possibility is higher; according to a plurality of the predetermined end points, determine the target end point of the obstacle, wherein the target end point includes the historical trajectory end point and the future reach end point; and, according to the target end point, determine the obstacle The estimated movement trajectory of the object from the end point of the historical trajectory to the end point in the future.

Optionally, the obstacle information includes historical position information of the obstacle; the map information includes lane line information, and the lane line information includes lane line adjacency relationship and lane line connection relationship; , determining a plurality of candidate end points of the obstacle, including: determining a basic lane matching the actual moving direction of the obstacle according to the historical position information; The lane line connection relationship of the lanes, expanding a plurality of candidate lanes around the basic lane; and determining a plurality of candidate end points according to the plurality of candidate lanes.

Optionally, the extracting the features of the multiple candidate endpoints according to the multiple candidate endpoints includes: using a multi-layer perceptron to determine the features of the multiple candidate endpoints according to the multiple candidate endpoints.

Optionally, the determining a plurality of candidate end points according to the plurality of candidate lanes includes: uniformly selecting a plurality of points on each of the candidate lanes at predetermined intervals as the plurality of candidate end points.

Optionally, the selecting a plurality of points uniformly on each of the candidate lanes at predetermined intervals as the plurality of candidate end points includes: along the direction of the lane line of the candidate lane, uniformly at predetermined intervals on Select a plurality of points on the lane centerline of each of the candidate lanes as a plurality of candidate end points; and/or, select a plurality of points uniformly on each of the candidate lanes at predetermined intervals as a plurality of points The candidate end points include: along the direction of the lane line of the candidate lane and the direction perpendicular to the direction of the lane line, uniformly select a plurality of points on each of the candidate lanes at predetermined intervals, as a plurality of points candidate endpoints.

Optionally, the information of the obstacle includes historical track information of the obstacle; the extracting the feature of the obstacle according to the information of the obstacle includes: using a convolutional neural network according to the historical track information to extract The obstacle feature, the time feature and spatial feature of the obstacle movement; and/or, the map information includes lane line information, drivable area information and semantic information; the map feature is extracted according to the map information, The method includes: extracting the map feature by using a graph neural network according to the lane line information, the drivable area information and the semantic information.

Optionally, the determining a plurality of predetermined endpoints from the plurality of candidate endpoints includes: judging one by one whether a plurality of the candidate endpoints will become the target endpoint; determining a plurality of candidate endpoints that may become the target endpoint and/or, said determining a plurality of predetermined endpoints from a plurality of said candidate endpoints includes: determining the probability that all candidate endpoints become the target endpoint, and selecting the most likely predetermined endpoint as the selected endpoint the target end point; and/or, determining the target end point of the obstacle according to a plurality of the predetermined end points includes: using a deep neural network to extract the characteristics of the predetermined end point; The map feature and the obstacle feature are used to optimize the predetermined end point and determine the target end point.

Optionally, the determining the estimated movement trajectory of the obstacle from the historical trajectory end point to the future reaching end point according to the target end point includes: using an optimization algorithm to determine the predicted motion trajectory according to the target end point. Estimating the motion trajectory, so that the estimated motion trajectory is smooth and conforms to the kinematics characteristics of the obstacle; and/or, according to the target end point, determining the distance from the end point of the historical trajectory to the future achievement of the obstacle The estimated trajectory of the destination includes: extracting the features of the target destination according to the destination destination; inputting the features of the destination destination, the map features and the obstacle features into a deep neural network to output the The estimated trajectory of the obstacle.

The present application provides an obstacle trajectory prediction system, including one or more processors, configured to implement any of the methods described above.

The present application provides a computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, the method described in any one of the above items is implemented.

In some embodiments, in the obstacle trajectory prediction method of the present application, the obtained obstacle information and the map information of the estimated movement of the obstacle are clear; the extracted features, map features and obstacle features of multiple candidate endpoints , is clear, and, according to the characteristics of candidate endpoints, map features, and obstacle characteristics, multiple predetermined endpoints are determined from multiple candidate endpoints, and the predetermined endpoints are also clear. Finally, according to the target endpoint, the history of obstacles can be determined The estimated trajectory from the end of the trajectory to the end in the future. This method can first determine the target end point for estimation, and then determine the estimated motion trajectory based on the historical trajectory end point and the future reached end point. The estimated motion trajectory is complete, so that the estimation process of the target end point is easier for developers to understand and easy to use The target end point explains the final estimated motion trajectory, avoiding completely black-box prediction results, which is conducive to later optimization, and facilitates timely grasp of the determination process of the estimated motion trajectory.

Description of drawings

Fig. 1 is a schematic flow chart of an embodiment of the obstacle trajectory prediction method provided by the present application;

FIG. 2 is a schematic flow chart of an embodiment of step 120 in the obstacle trajectory prediction method shown in FIG. 1;

FIG. 3 is a module block diagram of an embodiment of the obstacle trajectory prediction system provided by the present application.

Detailed ways

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numerals in different drawings refer to the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with aspects of one or more embodiments of the present specification as recited in the appended claims.

It should be noted that in other embodiments, the steps of the corresponding methods may not necessarily be performed in the order shown and described in this specification. In some other embodiments, the method may include more or less steps than those described in this specification. In addition, a single step described in this specification may be decomposed into multiple steps for description in other embodiments; multiple steps described in this specification may also be combined into a single step in other embodiments describe.

The method of the related technology is a black-box model, and its internal working mechanism is difficult for developers to understand. This is not conducive to the later optimization of the method, and it is not convenient to grasp the determination process of the estimated motion trajectory in time. In order to solve this technical problem, this In the obstacle trajectory prediction method provided in the embodiment of the application, the obtained obstacle information and the map information of the estimated movement of the obstacle are clear; the extracted features of multiple candidate endpoints, map features and obstacle features are Clear, and, according to the characteristics of the candidate endpoints, map features and obstacle characteristics, determine multiple predetermined endpoints from multiple candidate endpoints, the predetermined endpoints are also clear, and finally according to the target endpoint, the historical trajectory endpoint of the obstacle can be determined Estimated movement trajectory to reach the end point in the future. This method can first determine the target end point for estimation, and then determine the estimated motion trajectory based on the historical trajectory end point and the future reached end point. The estimated motion trajectory is complete, so that the estimation process of the target end point is easier for developers to understand and easy to use The target end point explains the final estimated motion trajectory, avoiding completely black-box prediction results, and is also convenient for other developers to understand, which is conducive to later optimization, and facilitates timely grasp of the determination process of the estimated motion trajectory.

The following firstly introduces the obstacle trajectory prediction method provided by the embodiment of the present application.

The obstacles in the obstacle trajectory prediction method provided in the embodiment of the present application may include moving objects. The mobile objects include, for example, people carrying mobile devices, one or more of non-motor vehicles and motor vehicles. Mobile devices may include cell phones and/or phone watches with information gathering capabilities. It is not limited here.

This obstacle trajectory prediction method can be applied to terminal equipment. The terminal device may include a vehicle, and the vehicle may include at least one self-driving vehicle. The at least one autonomous vehicle may include one or more of the current vehicle and other vehicles other than the current vehicle. If at least one self-driving vehicle can include at least one other vehicle, based on the current vehicle, the current vehicle uses the obstacle trajectory prediction method, and may encounter an obstacle containing at least one other vehicle during the movement of the current vehicle , can predict the trajectory of obstacles, and the current vehicle can better avoid or plan driving routes. The information of the current vehicle can be used to reflect the current vehicle's own situation, which includes the position information of the current vehicle and the motion state information of the current vehicle. In this way, the current position state of the vehicle can be better reflected.

FIG. 1 is a schematic flowchart of an embodiment of an obstacle trajectory prediction method provided by the present application.

As shown in FIG. 1 , the obstacle trajectory prediction method provided by the embodiment of the present application may include the following steps 110 to 160 : Step 110 , obtaining information about obstacles and map information of estimated movement of obstacles.

The information of the obstacle in this step 110 is used to reflect the situation of the obstacle itself, which includes historical position information of the obstacle and historical motion state information of the obstacle. The historical location information of the obstacle may include one or more of the historical GPS (Global Positioning System, GPS) coordinates of the obstacle, and the historical location of the obstacle relative to the current vehicle. The historical movement state information of the obstacle may include one or more of historical speed, historical acceleration, historical attitude angle, historical angular velocity, and historical angular acceleration. The historical position information of obstacles and the historical motion state information of obstacles are different from the information of the current vehicle in terms of time length. The historical position information of obstacles and the historical motion state information of obstacles are recorded before the current time. The current vehicle’s Information is obtained in real time.

Step 120, according to the map information, determine multiple candidate end points of the obstacle.

The map information is used to reflect the map information that obstacles may pass through, and includes one or more of lane line information, drivable area information, and semantic information. The lane marking information may include one or more of lane marking position and shape, lane marking adjacency relationship, lane marking connection relationship, and lane marking type. The lane-line adjacency relationship may refer to the left-right adjacent relationship, and the lane-line connection relationship may refer to the relationship between lanes extending forward and backward of the same lane. The lane marking type may refer to one or more of going straight and/or turning. Wherein, the lane may be composed of a lane center line, a lane left line, and a lane right line. Under ideal conditions, when the vehicle is driving straight, the trajectory of the vehicle center is relatively close to the centerline of the lane. When a vehicle is driving in the area formed by the middle of the left line of the lane and the right line of the lane, the vehicle can be said to belong to this lane. The drivable area may include one or more of a motor vehicle lane area, a non-motor vehicle lane area, and a road boundary. The semantic information may include one or more of intersection area, crosswalk area, and traffic restriction information. Traffic restriction information is used to indicate that obstacles comply with traffic rules, so that obstacles can be used to comply with traffic rules, and the trajectory of obstacles can be accurately predicted. The traffic restriction information may include one or more of traffic light information, speed limit information, and traffic restriction information.

FIG. 2 is a schematic flowchart of an embodiment of step 120 in the obstacle trajectory prediction method shown in FIG. 1 . As shown in FIG. 2 , step 120 includes the following steps 121 to 123 : step 121 , according to the historical position information, determine the basic lane matching the actual moving direction of the obstacle.

When there are many lanes in the intersection area and intersect with each other, obstacles may pass through multiple lanes. According to the historical position information, determine the lane that matches the actual movement direction of the obstacle as the basic lane, so that obstacles can be determined. The lane to which it belongs.

Step 122 , expand a plurality of candidate lanes around the basic lane according to the lane-line adjacency relationship of the basic lane and the lane-line connection relationship of the basic lane. Starting from the basic lane of the obstacle, multiple longer and more complete candidate lanes can be expanded to cover one or more motion requirements such as lane change, turning, and straight ahead of the obstacle. For example, if the obstacle is in the basic lane before the intersection area, the obstacle can either go straight or turn right, and two candidate lanes will be expanded. One candidate lane can be the lane that continues straight along the basic lane, and the other candidate lane can be The right turn lane continues along the base lane.

Step 123, determine a plurality of candidate end points according to the plurality of candidate lanes.

An embodiment of this step 123 may include selecting a plurality of points evenly on each candidate lane at a predetermined interval, as a plurality of candidate end points, the predetermined interval may be determined according to the accuracy of the estimated motion trajectory required by the user, and the estimated motion trajectory The greater the accuracy of the estimated motion trajectory, the smaller the predetermined interval, and the smaller the accuracy of the estimated motion trajectory, the larger the predetermined interval, which is not limited here. In this way, multiple points are uniformly selected on each candidate lane, and the determined candidate lane more accurately reflects the situation of the entire candidate lane, improving the accuracy of multiple candidate end points. Further, in this step 123, for each candidate lane in a plurality of candidate lanes, the candidate lane can be segmented to obtain a segmented lane; Select multiple points as multiple candidate endpoints. In this way, the amount of calculation for a segmented lane is less than the amount of calculation for the entire segment of candidate lanes, thereby improving processing efficiency.

There are multiple embodiments of selecting multiple points uniformly in each candidate lane at predetermined intervals as multiple candidate end points. In one embodiment, along the direction of the lane line of the candidate lane, uniformly select points at predetermined intervals at each Select multiple points on the lane centerline of each candidate lane as multiple candidate end points, so that the candidate end point is selected based on the centerline, and a better prediction effect of the estimated motion trajectory can be obtained while controlling the amount of calculation. . In another embodiment, along the direction of the lane line of the candidate lane and the direction perpendicular to the direction of the lane line, a plurality of points are uniformly selected on each candidate lane at predetermined intervals as the plurality of candidate end points. Such a dense selection of points in the candidate lane will increase the amount of calculation to a certain extent, but can obtain more complete and accurate prediction results. Among them, the distribution of multiple points selected in the direction of the lane line of the candidate lane and the direction perpendicular to the direction of the lane line can be in an array or a grid. In this way, the selected points are not only dense, but also relatively regular, so that the selected candidate end point It is more reasonable, and the estimated motion trajectory obtained is more reasonable.

Step 130, extract features of multiple candidate destinations according to multiple candidate destinations, extract map features according to map information, and extract obstacle features according to obstacle information.

Wherein, extracting features of multiple candidate endpoints according to multiple candidate endpoints in step 130 may include determining features of multiple candidate endpoints based on multiple candidate endpoints using a multi-layer perceptron. In this way, the multi-layer perceptron can be used independently to determine the features of multiple candidate endpoints, so that the features of multiple candidate endpoints can be better mined, so as to better extract the features of the target endpoint.

Extracting map features based on map information in step 130 may include extracting map features using a graph neural network based on lane line information, drivable area information, and semantic information, wherein the map features may include topological features between multiple candidate lanes, location and semantic features. In this way, the graph neural network can be used independently to visually determine the map features, which is convenient for understanding the map features, so that the map features can be better mined.

The extraction of obstacle features based on obstacle information in step 130 includes: extracting temporal and spatial features of obstacle movement by using a convolutional neural network based on historical trajectory information. In this way, the convolutional neural network can be used independently to extract the temporal and spatial features of the obstacle movement, so that the temporal and spatial characteristics of the obstacle movement can be better mined to facilitate the determination of the target end point. Further, this step 130 may also include a preprocessing operation on the multiple candidate destinations, map information, and obstacle information, wherein the preprocessing operation includes coordinate conversion and information encoding. In this way, it is convenient to remove some noise related to features, improve the accuracy of feature extraction, and then perform extraction of features, map features and obstacle features of multiple candidate endpoints.

Step 140, according to the characteristics of the candidate destinations, the map characteristics and the obstacle characteristics, determine a plurality of predetermined destinations from the plurality of candidate destinations. The probability of the target endpoint is higher. The predetermined end point may refer to the initially determined end point, and step 150 needs to be performed to determine whether the predetermined end point can become the final target end point.

This step 140 can be implemented in various embodiments. In some embodiments, it is judged one by one whether multiple candidate endpoints will become target endpoints; multiple candidate endpoints that may become target endpoints are determined as multiple predetermined endpoints. Setting multiple predetermined endpoints in this way can predict multiple estimated motion trajectories, covering a more complete trajectory space, and can also reduce the mutation of the estimated motion trajectory and ensure stability. At the same time, determining candidate endpoints one by one can improve the determination of candidate endpoints. accuracy. Specifically, a multi-layer perceptron and/or an attention algorithm may be used to determine whether multiple candidate endpoints will become target endpoints.

In other embodiments, the probability of all candidate endpoints becoming the target endpoint is determined, and the predetermined endpoint with the highest possibility is selected as the target endpoint. In this way, the passing probability can quickly determine the target end point, and the speed is faster. Specifically, a multilayer perceptron and/or an attention algorithm may be used to determine the probability that all candidate endpoints become target endpoints.

Step 150, determine the target end point of the obstacle according to multiple predetermined end points, wherein the target end point includes the end point of the historical trajectory and the end point reached in the future. In this way, the predetermined end point is optimized, and the position of the predetermined end point is adjusted, so that the coordinate point of the predetermined end point is more accurate as the end point of the estimated motion track.

An embodiment of this step 150 may include: using a deep neural network to extract features of the predetermined destination; optimizing the predetermined destination and determining the target destination according to the characteristic map features and obstacle characteristics of the predetermined destination. By adjusting the setting method of the end point, the generation of the predicted trajectory can be controlled to ensure the validity of the end point and the reliability of the estimated motion trajectory. The target end point can then be checked to ensure that the estimated trajectory conforms to vehicle kinematics, follows traffic regulations, has no collisions, etc.

Step 160, according to the target end point, determine the estimated movement track of the obstacle from the end point of the historical track to the end point in the future.

There are many implementations of this step 160. In one embodiment, according to the target end point, an optimization algorithm is used to determine the estimated motion trajectory, so that the estimated motion trajectory is smooth and conforms to the kinematic characteristics of the obstacle. Using the optimization algorithm in this way can detect the estimated motion trajectory and obtain a smooth estimated motion trajectory that conforms to the kinematic characteristics of obstacles, which can ensure the rationality of the estimated motion trajectory, thereby ensuring the safety of the current vehicle and reducing the occurrence of accidents possibility. In another embodiment, the features of the target end point are extracted according to the target end point; the feature of the target end point, the map feature and the obstacle feature are input into the deep neural network to output the estimated movement trajectory of the obstacle. In this way, the strong function fitting ability of the deep neural network is used, and at the same time, the artificial design of complex optimization algorithm constraints is avoided.

FIG. 3 is a module block diagram of an embodiment of an obstacle trajectory prediction system 300 provided by the present application. The obstacle trajectory prediction system 300 includes one or more processors 301 for implementing the obstacle trajectory prediction method described above.

In some embodiments, the obstacle trajectory prediction system 300 may include a computer-readable storage medium 309, which may store a program that can be invoked by the processor 301, and may include a non-volatile storage medium. In some embodiments, obstacle trajectory prediction system 300 may include memory 308 and interface 307 . In some embodiments, the obstacle trajectory prediction system 300 may also include other hardware according to actual applications.

The computer-readable storage medium 309 of the embodiment of the present application stores a program thereon, and when the program is executed by the processor 301, it is used to implement the obstacle trajectory prediction method described above.

This application may take the form of a computer program product embodied on one or more computer readable storage media 309 (including but not limited to disk storage, CD-ROM, optical storage, etc.) having program code embodied therein. Computer-readable storage media 309 includes both permanent and non-permanent, removable and non-removable media, and information storage may be implemented by any method or technology. Information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer readable storage media 309 include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), Read memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cartridge, tape magnetic disk storage or other magnetic storage device or any other non-transmission medium that may be used to store information that can be accessed by a computing device.

The above descriptions are only preferred embodiments of this specification, and are not intended to limit this specification. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of this specification shall be included in this specification. within the scope of protection.

It should also be noted that the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus comprising a set of elements includes not only those elements, but also includes Other elements not expressly listed, or elements inherent in the process, method, commodity, or apparatus are also included. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

Claims

A method for predicting obstacle trajectory, comprising:

Obtaining information about the obstacle and map information of the estimated movement of the obstacle;

determining a plurality of candidate end points of the obstacle according to the map information;

extracting features of a plurality of candidate endpoints according to the plurality of candidate endpoints, extracting map features according to the map information, and extracting the obstacle features according to the information of the obstacles;

According to the characteristics of the candidate destinations, the map characteristics and the obstacle characteristics, determine a plurality of predetermined destinations from the plurality of candidate destinations, and compare the plurality of predetermined destinations with those of the plurality of candidate destinations The rest of the candidate endpoints are more likely to become the target endpoints of the estimated trajectory;

determining a target end point of the obstacle according to a plurality of predetermined end points, wherein the target end point includes an end point of a historical trajectory and an end point reached in the future; and,

According to the target end point, an estimated movement track of the obstacle from the end point of the historical track to the end point reached in the future is determined.
The method of claim 1, wherein

The information of the obstacle includes historical position information of the obstacle;

The map information includes lane line information, and the lane line information includes lane line adjacency relationship and lane line connection relationship;

The determining multiple candidate end points of the obstacle according to the map information includes:

Determining a basic lane matching the actual movement direction of the obstacle according to the historical position information;

expanding a plurality of candidate lanes around the basic lane according to the lane line adjacency relationship of the basic lane and the lane connection relationship of the basic lane;

A plurality of candidate end points are determined according to the plurality of candidate lanes.
The trajectory prediction method according to claim 2, wherein said extracting features of a plurality of said candidate endpoints according to said plurality of said candidate endpoints comprises:

According to the plurality of candidate endpoints, a multi-layer perceptron is used to determine the characteristics of the plurality of candidate endpoints.
The trajectory prediction method according to claim 2, wherein said determining a plurality of said candidate end points according to said plurality of said candidate lanes comprises:

A plurality of points are evenly selected on each of the candidate lanes at predetermined intervals as the plurality of candidate end points.
The trajectory prediction method according to claim 4, wherein said uniformly selecting a plurality of points on each of said candidate lanes at predetermined intervals as a plurality of said candidate end points includes any one or more of the following item:

along the direction of the lane line of the candidate lane, uniformly select a plurality of points on the lane centerline of each of the candidate lanes at predetermined intervals as a plurality of candidate end points;

Along the direction of the lane line of the candidate lane and a direction perpendicular to the direction of the lane line, a plurality of points are uniformly selected on each of the candidate lanes at predetermined intervals as the plurality of candidate end points.
trajectory prediction method as claimed in claim 1, is characterized in that,

The information of the obstacle includes historical track information of the obstacle;

The extracting the obstacle feature according to the information of the obstacle includes: extracting the obstacle feature, the time feature and the space feature of the obstacle movement according to the historical track information by using a convolutional neural network.
The trajectory prediction method according to claim 1 or 6, wherein,

The map information includes lane line information, drivable area information and semantic information;

The extracting map features according to the map information includes: extracting the map features by using a graph neural network according to the lane line information, the drivable area information and the semantic information.
The trajectory prediction method according to claim 1, wherein said determining a plurality of predetermined endpoints from a plurality of said candidate endpoints comprises at least one of the following:

judging whether multiple candidate endpoints will become the target endpoint one by one; determining multiple candidate endpoints that may become target endpoints as multiple predetermined endpoints;

determining the probability that all candidate endpoints become the target endpoint, and selecting the predetermined endpoint with the greatest possibility as the target endpoint;

Using a deep neural network to extract the features of the predetermined destination; optimizing the predetermined destination and determining the target destination according to the characteristics of the predetermined destination, the map features and the characteristics of the obstacles.
The trajectory prediction method according to claim 1, wherein said determining the estimated movement trajectory of said obstacle from said historical trajectory end point to said future reaching end point according to said target end point comprises at least the following one item:

According to the target end point, use an optimization algorithm to determine the estimated motion trajectory, so that the estimated motion trajectory is smooth and conforms to the kinematic characteristics of the obstacle;

Extracting the features of the target end point according to the target end point; inputting the feature of the target end point, the map feature and the obstacle feature into a deep neural network to output the estimated movement trajectory of the obstacle.
A trajectory prediction system, characterized by comprising one or more processors for implementing the trajectory prediction method according to any one of claims 1-9.
A computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, the trajectory prediction method according to any one of claims 1 to 9 is realized.