Trajectory prediction method, device, electronic device and readable storage medium

By combining real-time and historical data of obstacles, multiple effective prediction trajectories are determined and optimal trajectories are selected, which solves the problem of poor trajectory prediction reliability in the prior art and improves the prediction accuracy of unmanned vehicles.

CN118597186BActive Publication Date: 2024-12-06HIGER

Patent Information

Application Number
CN202410609722.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-16
Publication Date
2024-12-06
Estimated Expiration
2044-05-16

AI Technical Summary

Technical Problem

In the prior art, only real-time driving data of obstacles is used to predict the trajectory, and it is difficult to predict the driving situation of obstacles at intersections, resulting in poor trajectory prediction reliability and increasing the risk of unmanned driving.

Method used

By obtaining real-time driving data, historical driving data and historical prediction data of obstacles, multiple effective prediction trajectories on multiple target lanes, and selecting the optimal prediction trajectory based on the confidence of each trajectory, ensuring that the prediction trajectory is in line with complexity Road environment.

Benefits of technology

Improve the accuracy and reliability of trajectory prediction, ensuring that obstacle prediction trajectories that meet the actual situation are generated in complex road environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118597186B_ABST
    Figure CN118597186B_ABST
Patent Text Reader

Abstract

The present application discloses a trajectory prediction method, device, electronic device and readable storage medium. The method of the present application obtains the real-time driving data of the obstacle, the historical driving data and historical prediction data of the obstacle in the previous round of prediction actions; determines multiple valid prediction trajectories of the obstacle on multiple target lanes based on the real-time driving data, historical driving data and historical prediction data of the obstacle; obtains the trajectory confidence corresponding to each valid prediction trajectory; among multiple valid prediction trajectories, determines the valid prediction trajectory corresponding to the maximum trajectory confidence as the optimal prediction trajectory of the obstacle. The optimal prediction trajectory of the obstacle is determined by utilizing the validity and trajectory confidence of multiple prediction trajectories of the obstacle under various intention types in a complex road environment, ensuring that the predicted moving trajectory conforms to the complex road conditions and improving the accuracy and reliability of the predicted trajectory.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of trajectory planning, and in particular to a trajectory prediction method, device, electronic device and readable storage medium. Background Art

[0002] At present, obstacle trajectory prediction is an important technology in the field of unmanned driving. In the existing technology, sensors are used to collect driving data such as the position, heading, and speed of obstacles to generate predicted trajectories. However, only predicting the trajectory with the real-time driving data of obstacles easily ignores the impact of road structures such as intersections on the driving trajectory, and it is difficult to predict the driving conditions of obstacles at intersections, resulting in poor reliability of trajectory prediction, which in turn increases the risk of unmanned driving. Summary of the invention

[0003] In view of this, the present application provides a trajectory prediction method, device, electronic device and readable storage medium, the main purpose of which is to solve the technical problem in the prior art that the trajectory is predicted only by the real-time driving data of the obstacle, which makes it difficult to predict the driving conditions of the obstacle at the intersection, resulting in poor trajectory prediction reliability.

[0004] According to a first aspect of the present application, a trajectory prediction method is provided, the method comprising:

[0005] Obtain the real-time driving data of the obstacle, the historical driving data of the obstacle in the previous round of prediction actions, and the historical prediction data;

[0006] Determine multiple valid predicted trajectories of the obstacle on multiple target lanes based on the real-time driving data, historical driving data and historical prediction data of the obstacle;

[0007] Get the trajectory confidence corresponding to each valid predicted trajectory;

[0008] Among multiple valid predicted trajectories, the valid predicted trajectory corresponding to the maximum trajectory confidence is determined as the optimal predicted trajectory of the obstacle.

[0009] Optionally, the step of obtaining the real-time driving data of the obstacle, the historical driving data of the obstacle in the previous round of prediction actions, and the historical prediction data specifically includes:

[0010] Obtain the location information of the moving vehicle on the high-precision map;

[0011] With the location of the moving vehicle as the origin and the preset distance as the radius, a target circular area is generated on the high-precision map;

[0012] Obtaining multiple curvature values ​​of multiple target lanes within the target circular area;

[0013] Obtaining the identity identification, first position coordinate data and real-time driving data of obstacles on multiple target lanes, wherein the real-time driving data includes: real-time driving speed, real-time acceleration and real-time heading angle;

[0014] According to the identity identifier, the historical driving data and the historical prediction data of the obstacle in the previous round of prediction actions are obtained from the multiple stored data corresponding to the previous round of prediction actions;

[0015] The historical driving data includes the historical driving speed, historical acceleration and historical heading angle of the obstacle in the previous round of prediction actions; the historical prediction data includes multiple historical prediction trajectories of the obstacle in the previous round of prediction actions, and the historical trajectory confidence of each historical prediction trajectory.

[0016] Optionally, the step of determining multiple valid predicted trajectories of the obstacle on multiple target lanes according to the real-time driving data, historical driving data and historical prediction data of the obstacle specifically includes:

[0017] The real-time driving data, historical driving data and historical prediction data of the obstacle are input into a preset trajectory prediction model, and multiple predicted trajectories of the obstacle on multiple target lanes and the prediction confidence corresponding to each predicted trajectory are output;

[0018] Get the intent type of each predicted trajectory;

[0019] For any predicted trajectory, if the intention type is a straight-ahead type, the predicted trajectory is determined to be a valid trajectory;

[0020] If the intention type is a turning type or a U-turn type, obtain the minimum deceleration distance of the obstacle in the predicted trajectory, the Euclidean distance between the obstacle and the nearest turning point on the predicted trajectory, and the maximum longitudinal deceleration of the obstacle before turning;

[0021] If the Euclidean distance between the obstacle and the turning point is less than the minimum deceleration distance and / or the real-time acceleration of the obstacle is greater than the maximum longitudinal deceleration of the obstacle before turning, the predicted trajectory is determined to be an invalid predicted trajectory;

[0022] If the Euclidean distance between the obstacle and the turning point is greater than or equal to the minimum deceleration distance, and the real-time acceleration of the obstacle is less than or equal to the maximum longitudinal deceleration of the obstacle before turning, the predicted trajectory is determined to be a valid predicted trajectory.

[0023] Optionally, the step of obtaining a minimum deceleration distance of an obstacle in the predicted trajectory, a Euclidean distance between the obstacle and the nearest turning point on the predicted trajectory, and a maximum longitudinal deceleration of the obstacle before turning specifically includes:

[0024] Obtain the second position coordinate of the turning point on the predicted trajectory that is closest to the obstacle;

[0025] Determine the Euclidean distance between the obstacle and the turning point according to the first position coordinate and the second position coordinate;

[0026] Obtain the maximum centripetal acceleration of the obstacle at the turn on the predicted trajectory and the maximum longitudinal deceleration before the turn;

[0027] The turning speed and minimum deceleration distance of the obstacle on the predicted trajectory are determined based on the maximum centripetal acceleration at the turn, the curvature value of the lane where the predicted trajectory is located, the real-time driving speed, and the maximum longitudinal deceleration before the turn.

[0028] Optionally, the step of determining the turning speed and minimum deceleration distance of the obstacle on the predicted trajectory according to the maximum centripetal acceleration at the turning, the curvature value of the lane where the predicted trajectory is located, the real-time driving speed, and the maximum longitudinal deceleration before the turn specifically includes:

[0029] The turning speed of the obstacle on the predicted trajectory is calculated using the turning centripetal acceleration and the curvature value of the lane where the predicted trajectory is located;

[0030] If the turning speed is greater than the real-time speed of the obstacle, the real-time speed is used as the turning speed of the obstacle on the predicted trajectory;

[0031] The minimum deceleration distance of the obstacle on the predicted trajectory is calculated using the obstacle's turning speed, real-time driving speed, and maximum longitudinal deceleration before turning.

[0032] Optionally, the step of obtaining the trajectory confidence corresponding to each valid predicted trajectory specifically includes:

[0033] For any valid predicted trajectory, obtain the preset heading angle deviation calculation weight value, and effectively predict the target heading angle of the target position corresponding to the preset time on the trajectory;

[0034] The weight value is calculated using the obstacle's real-time heading angle, historical heading angle, target heading angle and preset heading angle deviation to calculate the heading angle deviation confidence level corresponding to the effective predicted trajectory.

[0035] Get the preset speed deviation calculation weight value;

[0036] The speed at which the obstacle reaches the turning point is calculated using the obstacle's real-time speed, historical speed, real-time acceleration, historical acceleration, and the Euclidean distance between the obstacle and the turning point.

[0037] The speed deviation confidence corresponding to the effective predicted trajectory is calculated by using the turning speed of the obstacle on the effective predicted trajectory, the driving speed of the obstacle at the turning point, and the preset speed deviation to calculate the weight value;

[0038] Obtain the preset confidence calculation weight value of the obstacle in the previous round of prediction action;

[0039] Obtain multiple similarities between the effective prediction trajectory and multiple historical prediction trajectories;

[0040] Determine the target historical prediction trajectory with the highest similarity among multiple historical prediction trajectories;

[0041] The continuous prediction confidence corresponding to the effective prediction trajectory is calculated by using the prediction confidence corresponding to the effective prediction trajectory, the historical trajectory confidence of the target historical prediction trajectory and the preset confidence calculation weight value;

[0042] The heading angle deviation confidence, velocity deviation confidence and continuous prediction confidence are added together to obtain the trajectory confidence of the effective predicted trajectory.

[0043] Optionally, according to the identity identifier, before obtaining the historical driving data and historical prediction data of the obstacle in the previous round of prediction actions from the plurality of stored data corresponding to the previous round of prediction actions, the method further includes:

[0044] According to the identity of the obstacle, determining whether historical data of the obstacle is stored in the plurality of stored data;

[0045] If the historical data of the obstacle is not stored in the multiple stored data, the real-time driving speed, real-time acceleration and real-time heading angle of the obstacle are input into a preset trajectory prediction model, and multiple predicted trajectories and a prediction confidence of each predicted trajectory are output;

[0046] Among multiple predicted trajectories, the predicted trajectory corresponding to the maximum prediction confidence is determined as the optimal predicted trajectory of the obstacle.

[0047] According to a second aspect of the present application, a trajectory prediction device is provided, the device comprising:

[0048] A first acquisition module is used to acquire real-time driving data of the obstacle, historical driving data of the obstacle in the previous round of prediction actions, and historical prediction data;

[0049] A first determination module, used to determine multiple valid predicted trajectories of the obstacle on multiple target lanes according to the real-time driving data, historical driving data and historical prediction data of the obstacle;

[0050] The second acquisition module is used to obtain the trajectory confidence corresponding to each valid predicted trajectory;

[0051] The second determination module is used to determine, among multiple valid predicted trajectories, a valid predicted trajectory corresponding to a maximum trajectory confidence as an optimal predicted trajectory of the obstacle.

[0052] According to a third aspect of the present application, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the methods described in the first aspect when executing the computer program.

[0053] According to a fourth aspect of the present application, a readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the methods in the first aspect are implemented.

[0054] By means of the above technical solution, the present application provides a trajectory prediction method, device, electronic device and readable storage medium. Specifically, after monitoring an obstacle vehicle near an autonomous driving vehicle, the real-time driving data of the obstacle is collected, and the historical driving data and historical trajectory data of the obstacle in the previous round of prediction actions are combined to generate multiple predicted trajectories of the obstacle in this round of prediction actions, and the validity of the multiple predicted trajectories is verified to obtain multiple valid predicted trajectories. Thereafter, based on the intention type corresponding to each valid predicted trajectory, the trajectory confidence of each valid predicted trajectory is determined, and finally the valid predicted trajectory with the largest trajectory confidence is used as the optimal predicted trajectory of this round of prediction actions. Compared with the prediction method of the prior art that only uses the real-time driving data of the obstacle for trajectory prediction, the present application obtains the validity and trajectory confidence of multiple predicted trajectories of the obstacle under multiple intention types in a complex road environment, and then determines the optimal predicted trajectory of the obstacle based on the validity and trajectory confidence of the multiple predicted trajectories, ensuring that the predicted moving trajectory conforms to the complex road conditions and improving the accuracy and reliability of the predicted trajectory.

[0055] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0057] Figure 1 A schematic diagram of a trajectory prediction method provided in an embodiment of the present application is shown;

[0058] Figure 2 A schematic diagram of the topological structure of multiple target lanes in a target circular area near a moving vehicle provided in an embodiment of the present application is shown;

[0059] Figure 3 One of the schematic diagrams of the structure of an obstacle on a high-precision map provided by an embodiment of the present application is shown;

[0060] Figure 4 The second schematic diagram of the structure of an obstacle on a high-precision map provided by an embodiment of the present application is shown;

[0061] Figure 5 A schematic diagram of the structure of a trajectory prediction system provided in an embodiment of the present application is shown;

[0062] Figure 6 A schematic diagram of the structure of a trajectory prediction device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0063] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0064] The present application embodiment provides a trajectory prediction method, such as Figure 1 As shown, the method includes:

[0065] S101, obtaining real-time driving data of an obstacle, historical driving data of the obstacle in a previous round of prediction actions, and historical prediction data.

[0066] The trajectory prediction method provided in the embodiment of the present application is mainly used in the trajectory prediction scenario of dynamic obstacles on the road monitored by the vehicle during unmanned driving. The execution subject of the embodiment of the present application is a device or equipment capable of trajectory prediction, which can be specifically set on the server side, and the server automatically predicts the trajectory of obstacles during driving.

[0067] Specifically, obstacles are other vehicle participants around the autonomous vehicle, such as motor vehicles or non-motor vehicles near the autonomous vehicle. The positions and speeds of these obstacles are constantly changing, and their appearance has an important impact on the driving trajectory and decision-making of the unmanned vehicle. Therefore, it is necessary to monitor and predict the behavior of obstacles in real time to ensure safe driving and make appropriate driving decisions. In this round of trajectory prediction, when obstacles are detected near the autonomous vehicle, sensors are used to collect real-time driving data of the obstacles.

[0068] Furthermore, the obstacle's driving trajectory is usually predicted at fixed time intervals (such as 100 milliseconds). In order to better analyze the obstacle's past driving patterns and trajectory changes, and thus better understand the obstacle's behavior and possible driving routes, the obstacle's historical driving data and historical prediction data in the previous round of prediction actions are obtained and combined with the obstacle's real-time driving data to more accurately predict the obstacle's moving path.

[0069] Optionally, the obstacle includes but is not limited to any one of a non-motorized vehicle and a vehicle traveling on the road.

[0070] S102: Determine multiple valid predicted trajectories of the obstacle on multiple target lanes according to the real-time driving data, historical driving data, and historical prediction data of the obstacle.

[0071] In this step, the real-time driving data of the obstacle, as well as the historical driving data and historical prediction data of the obstacle in the previous round of prediction actions are used as basic data to predict multiple moving trajectories of the obstacle on multiple target lanes. However, there are many interferences and constraints in the actual road environment, especially at turning intersections, where the complexity of the road environment is particularly prominent. The predicted trajectory of the obstacle requires accurate judgment and decision-making in these complex scenarios. Therefore, it is necessary to verify the validity of multiple predicted trajectories based on complex road environment data to verify whether the predicted trajectory conforms to the actual road conditions, so as to screen out effective predicted trajectories that ensure the driving safety of obstacles.

[0072] S103: Obtain the trajectory confidence corresponding to each valid predicted trajectory.

[0073] In this step, during the trajectory prediction process, due to the complex road environment, multiple valid predicted trajectories of obstacles contain multiple intent categories, such as keep going straight, turn left, turn right, etc. In order to improve the accuracy and adaptability of the final predicted trajectory of the obstacle in complex road conditions, the trajectory confidence of each valid predicted trajectory is obtained respectively. The higher the trajectory confidence, the more reliable the prediction of the valid predicted trajectory under its corresponding intent category.

[0074] S104. Determine, among multiple valid predicted trajectories, a valid predicted trajectory corresponding to a maximum trajectory confidence as an optimal predicted trajectory of the obstacle.

[0075] In this step, after obtaining the trajectory confidence corresponding to each valid predicted trajectory, multiple valid predicted trajectories are sorted in descending order of trajectory confidence, and then the valid predicted trajectory with the largest trajectory confidence is determined as the optimal predicted trajectory of the obstacle in this round of prediction operation.

[0076] The trajectory prediction method provided in the embodiment of the present application, after monitoring the obstacle vehicle near the autonomous driving vehicle, collects the real-time driving data of the obstacle, combines the historical driving data and historical trajectory data of the obstacle in the previous round of prediction action, generates multiple predicted trajectories of the obstacle in this round of prediction action, verifies the validity of the multiple predicted trajectories, and obtains multiple valid predicted trajectories. Thereafter, based on the intention type corresponding to each valid predicted trajectory, the trajectory confidence of each valid predicted trajectory is determined, and finally the valid predicted trajectory with the largest trajectory confidence is used as the optimal predicted trajectory of this round of prediction action. Compared with the prediction method of the prior art that only uses the real-time driving data of the obstacle for trajectory prediction, the present application obtains the validity and trajectory confidence of multiple predicted trajectories of the obstacle under multiple intention types in a complex road environment, and then determines the optimal predicted trajectory of the obstacle based on the validity and trajectory confidence of the multiple predicted trajectories, ensures that the predicted moving trajectory conforms to the complex road conditions, and improves the accuracy and reliability of the predicted trajectory.

[0077] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, the embodiment of the present application provides another trajectory prediction method, which includes:

[0078] S201. Obtain the location information of the moving vehicle on the high-precision map.

[0079] S202. Generate a target circular area on a high-precision map with the location of the moving vehicle as the origin and a preset distance as the radius.

[0080] In steps S201 and S202, the driving vehicle is an autonomous driving vehicle. The autonomous driving vehicle perceives other traffic participants on the road (such as motor vehicles and non-motor vehicles) and accurately predicts the movement trajectories of the traffic participants, which is an important prerequisite for the autonomous driving vehicle to plan a safe, comfortable and efficient driving trajectory.

[0081] Furthermore, in the process of predicting the trajectory of obstacles near the moving vehicle, the position information of the moving autonomous vehicle on the high-precision map is obtained in real time, and a circle is drawn on the high-precision map with the position of the moving vehicle as the origin and the preset search distance (preset distance) as the radius, forming a target circular area centered on the moving vehicle. By taking the moving vehicle as the reference point and the lane area where the moving vehicle is located as the reference range for monitoring obstacles, more attention can be paid to obstacles that may appear around the moving vehicle, avoiding misjudgment or omission caused by too large or too small a range, and helping to more accurately locate and identify obstacles around the moving vehicle.

[0082] Optionally, the range of the preset distance for constructing the target circular area can be set according to the specific road environment where the autonomous driving vehicle is located. If the road environment is complex, such as dense intersections and heavy traffic, then the preset distance should be large enough to ensure that the autonomous driving vehicle has enough time and space to respond and avoid potential collisions or dangerous situations.

[0083] S203: Acquire multiple curvature values ​​of multiple target lanes within the target circular area.

[0084] In this step, during the obstacle trajectory prediction process, the curvature value of the lane helps to better understand the direction and shape of the current lane, and thus more accurately plan the movement path of the obstacle. Therefore, after constructing the target circular area near the moving vehicle, multiple target lanes contained in the target circular area are determined on the high-precision map, and the curvature value of each target lane is obtained.

[0085] In actual application scenarios, while obtaining the curvature value of each vehicle, the road topology of multiple target lanes in the target circular area can also be obtained, such as Figure 2 The figure shows the topological structure of multiple target lanes in the target circular area near the moving vehicle, where the target circular area includes lanes 101 to 109, and the front-to-back order, left-to-right adjacent relationship, and lane length, lane width, lane curvature, and lane driving direction of each target lane are obtained. This helps to understand important information such as intersections and divergences in multiple target lanes, so as to plan a safe and efficient driving path.

[0086] S204, obtaining the identity identification, first position coordinate data and real-time driving data of obstacles on multiple target lanes, wherein the real-time driving data includes: real-time driving speed, real-time acceleration and real-time heading angle.

[0087] In this step, the vehicle driving conditions on multiple target lanes within the target circular area are monitored in real time. If other vehicles besides the driving vehicle are detected on multiple target lanes, the vehicle is determined to be an obstacle. Obtaining the identity of the obstacles monitored in multiple target lanes facilitates accurate identification of the same obstacle in the previous round of prediction actions stored. Obtaining the first position coordinate data of the obstacle on the high-precision map and the real-time driving data of the obstacle, including real-time driving data, real-time acceleration, and real-time heading angle, helps predict the future trajectory of the obstacle.

[0088] Optionally, the number of obstacles monitored on the multiple target lanes in the target circular area may be one or more. If the number of obstacles is multiple, the optimal driving trajectory of each obstacle is predicted separately.

[0089] In actual application scenarios, when the sensor detects in real time that there is an obstacle on any target lane within the target circular area, the UTM (UNIVERSAL TRANSVERSE MERCARTOR GRID SYSTEM) coordinate system is used to project obstacle 1 onto the high-precision map. At the same time, the sensor is used to obtain the license plate information of the obstacle, and the identity of the obstacle, obstacle 1, is determined based on the license plate information, and the first position coordinates of obstacle 1 on the high-precision map and the real-time driving data of obstacle 1 are obtained.

[0090] S205 . According to the identity identifier, obtain historical driving data and historical prediction data of the obstacle in the previous round of prediction actions from a plurality of stored data corresponding to the previous round of prediction actions.

[0091] In this step, according to the identity of the obstacle, the historical driving data and historical prediction data of the obstacle in the previous round of prediction actions are retrieved from the multiple stored data of the previous round of prediction actions. Specifically, the historical driving data includes the historical driving speed, historical acceleration and historical heading angle of the obstacle in the previous round of prediction actions; the historical prediction data includes multiple historical prediction trajectories of the obstacle in the previous round of prediction actions, and the historical trajectory confidence of each historical prediction trajectory. The movement of a vehicle usually has a certain degree of continuity and inertia. By retrieving the driving data and prediction data of the obstacle in the previous round of prediction actions to ensure the continuity of the prediction trajectory, the movement trend and possible changes of the obstacle can be judged, thereby more accurately predicting the future trajectory of the obstacle.

[0092] Optionally, before retrieving the historical driving data and historical prediction data of the obstacle, the obstacle is matched with the historical obstacles monitored in the previous round of prediction actions according to the obstacle's identity to determine whether the obstacle was monitored in the previous round of prediction actions. If the obstacle was not monitored in the previous round of prediction actions, it means that the obstacle is being monitored for the first time, that is, the multiple stored data do not contain the historical data of the obstacle. At this time, the real-time driving data, real-time acceleration and real-time heading angle of the obstacle monitored for the first time are input into a preset trajectory prediction model, and multiple predicted trajectories and the prediction confidence of each predicted trajectory are output, and the predicted trajectory with the largest prediction confidence is used as the optimal predicted trajectory of the obstacle in this round of prediction actions.

[0093] Optionally, a large amount of driving data (such as speed, acceleration, heading angle, etc.) of obstacle vehicles for autonomous driving vehicles in real road environments, as well as the driving trajectory of obstacle vehicles are collected in advance, the data set is divided into a training set, a validation set, and a test set, the neural network model is trained, and a trained preset trajectory prediction model is generated.

[0094] S206: Determine multiple valid predicted trajectories of the obstacle on multiple target lanes according to the real-time driving data, historical driving data, and historical prediction data of the obstacle.

[0095] This step and Figure 1 The method of step S102 shown is the same and will not be repeated here.

[0096] In an embodiment of the present application, optionally, in step S206, multiple valid predicted trajectories of the obstacle on multiple target lanes are determined based on the real-time driving data, historical driving data and historical prediction data of the obstacle, specifically including: inputting the real-time driving data, historical driving data and historical prediction data of the obstacle into a preset trajectory prediction model, outputting multiple predicted trajectories of the obstacle on multiple target lanes, and the prediction confidence corresponding to each predicted trajectory; obtaining the intention type of each predicted trajectory; for any predicted trajectory, if the intention type is a straight type, determining the predicted trajectory to be a valid predicted trajectory; if the intention type is a turn, The predicted trajectory is determined as follows: if the Euclidean distance between the obstacle and the turning point is less than the minimum deceleration distance and / or the real-time acceleration of the obstacle is greater than the maximum longitudinal deceleration of the obstacle before the turn, the predicted trajectory is determined to be an invalid predicted trajectory; if the Euclidean distance between the obstacle and the turning point is greater than or equal to the minimum deceleration distance, and the real-time acceleration of the obstacle is less than or equal to the maximum longitudinal deceleration of the obstacle before the turn, the predicted trajectory is determined to be a valid predicted trajectory.

[0097] In this embodiment, the preset trajectory prediction model is a pre-trained deep learning model for predicting the movement trajectory of the obstacle. Specifically, the real-time driving data, historical driving data and historical prediction data of the obstacle are input into the preset trajectory prediction model, and multiple predicted trajectories of the obstacle on multiple target lanes and the prediction confidence of each predicted trajectory are output.

[0098] Furthermore, the predicted multiple driving trajectories contain multiple intention types, such as straight, turning, U-turn, etc. The credibility of the predicted trajectories varies depending on the intention type of the predicted trajectories. If the intention type of the predicted trajectory of the obstacle is a straight type, that is, the obstacle appears to move in a straight line consistent with the current driving direction, it usually has a higher confidence level; if the intention type of the predicted trajectory is a turning type or a U-turn type, that is, when the obstacle appears to turn or U-turn, the driving trajectory of the obstacle usually involves more uncertain factors, such as the minimum deceleration distance between the obstacle and the intersection, the turning speed of the obstacle, etc. Therefore, it is necessary to verify whether the predicted trajectory is valid.

[0099] For any predicted trajectory, if the intention type of the predicted trajectory is a straight-ahead type, the predicted trajectory is confirmed to be a valid predicted trajectory. If the intention type of the predicted trajectory is a turning type or a U-turn type, the minimum deceleration distance of the obstacle on the predicted trajectory, the Euclidean distance between the obstacle and the nearest turning point on the predicted trajectory, and the maximum longitudinal deceleration of the obstacle before turning are obtained. These data can be used to evaluate the safety of the obstacle's turning behavior to determine whether the predicted trajectory of the obstacle has a collision risk. Specifically, if the Euclidean distance between the obstacle and the nearest turning point is less than the minimum deceleration distance of the obstacle on the predicted trajectory, it means that the deceleration distance is too small or the obstacle is too close to the turning point, and the obstacle does not have enough space to turn safely. Therefore, the predicted trajectory is determined to be an invalid predicted trajectory. Furthermore, if the real-time acceleration of the obstacle is greater than the maximum longitudinal deceleration of the obstacle before turning, it means that the obstacle is approaching the turning point at a faster speed, increasing the risk of collision with the obstacle, especially if there is congestion at the turning intersection. Therefore, the predicted trajectory is determined to be an invalid predicted trajectory.

[0100] Furthermore, if the Euclidean distance between the obstacle and the nearest turning point is greater than or equal to the minimum deceleration distance of the obstacle on the predicted trajectory, and the real-time acceleration of the obstacle is less than or equal to the maximum longitudinal deceleration of the obstacle before turning, it indicates that the obstacle is traveling at a safe speed and there is enough space for the obstacle to decelerate or avoid, and it is unlikely that an unexpected situation will occur when turning, then the predicted trajectory is confirmed to be a valid predicted trajectory.

[0101] In the above manner, the validity of the predicted trajectory with turning intention is detected to provide more accurate and reliable decision support and ensure the accuracy of the predicted trajectory.

[0102] In an embodiment of the present application, optionally, the minimum deceleration distance of the obstacle in the predicted trajectory, the Euclidean distance between the obstacle and the nearest turning point on the predicted trajectory, and the maximum longitudinal deceleration of the obstacle before turning are obtained, specifically including: obtaining the second position coordinates of the turning point on the predicted trajectory that is closest to the obstacle; determining the Euclidean distance between the obstacle and the turning point based on the first position coordinates and the second position coordinates; obtaining the maximum centripetal acceleration of the obstacle at the turn on the predicted trajectory and the maximum longitudinal deceleration before turning; determining the turning speed and the minimum deceleration distance of the obstacle on the predicted trajectory based on the maximum centripetal acceleration at the turn, the curvature value of the lane where the predicted trajectory is located, the real-time driving speed, and the maximum longitudinal deceleration before turning.

[0103] In this embodiment, on the high-precision map, the intersection closest to the obstacle on the predicted trajectory is determined, that is, the turning point closest to the obstacle, and the second position coordinates of the turning point are obtained. The Euclidean distance between the obstacle and the turning point is calculated using the first position coordinates of the obstacle and the second position coordinates of the turning point.

[0104] Optionally, the Euclidean distance between the obstacle and the turning point:

[0105]

[0106] Wherein, the above Dc is the Euclidean distance; the above (X c , Y c ) is the first position coordinate of the obstacle; the above (X r , Y r ) is the second position coordinate of the turning point.

[0107] Furthermore, the maximum centripetal acceleration of the obstacle at the turn on the predicted trajectory and the maximum longitudinal deceleration before the turn are obtained. The turning speed and the minimum deceleration distance of the obstacle on the predicted trajectory are calculated based on the maximum centripetal acceleration at the turn, the curvature value of the lane where the predicted trajectory is located, the real-time driving speed, and the maximum longitudinal deceleration before the turn.

[0108] In an embodiment of the present application, optionally, the turning speed and minimum deceleration distance of the obstacle on the predicted trajectory are determined according to the maximum centripetal acceleration at the turn, the curvature value of the lane where the predicted trajectory is located, the real-time driving speed, and the maximum longitudinal deceleration before the turn, specifically including: calculating the turning speed of the obstacle on the predicted trajectory using the turning centripetal acceleration and the curvature value of the lane where the predicted trajectory is located; if the turning speed is greater than the real-time driving speed of the obstacle, using the real-time driving speed as the turning speed of the obstacle on the predicted trajectory; calculating the minimum deceleration distance of the obstacle on the predicted trajectory using the turning speed of the obstacle, the real-time driving speed, and the maximum longitudinal deceleration before the turn.

[0109] In this embodiment, the number of multiple effective predicted trajectories of obstacles is set to N. For any effective predicted trajectory, the turning speed of the obstacle is calculated using the maximum centripetal acceleration of the obstacle turning and the curvature of the lane where the predicted trajectory is located:

[0110]

[0111] Among them, the above V curveI is the turning speed of the obstacle in the Ith valid predicted trajectory, I∈1,2,...,N; the above A curve is the maximum centripetal acceleration of the obstacle at the turn on the predicted trajectory; the above K riis the curvature value of the lane corresponding to the i-th valid predicted trajectory, i=I, i∈1,2,...,N.

[0112] Optionally, the calculated turning speed of the obstacle on the predicted trajectory is compared with the real-time driving speed of the obstacle. If the turning speed of the obstacle on the predicted trajectory is greater than the real-time driving speed of the obstacle, the real-time driving speed of the obstacle is used as the turning speed of the obstacle on the predicted trajectory.

[0113] Furthermore, the minimum deceleration distance of the obstacle on the predicted trajectory is calculated using the calculated turning speed of the obstacle on the predicted trajectory, the real-time driving speed of the obstacle, and the preset maximum longitudinal deceleration of the obstacle before turning:

[0114]

[0115] Among them, the above D decI is the minimum deceleration distance of the obstacle on the Ith effective predicted trajectory; the above A dec is the maximum longitudinal deceleration of the obstacle before turning.

[0116] S207: Obtain the trajectory confidence corresponding to each valid predicted trajectory.

[0117] This step and Figure 1 The method of step S103 shown is the same and will not be repeated here.

[0118] In the embodiment of the present application, optionally, in step S207, the trajectory confidence corresponding to each valid predicted trajectory is obtained, specifically including: for any valid predicted trajectory, obtaining a preset heading angle deviation calculation weight value, and a target heading angle of the target position corresponding to the preset time on the effective predicted trajectory; using the obstacle's real-time heading angle, historical heading angle, target heading angle and preset heading angle deviation calculation weight value to calculate the heading angle deviation confidence corresponding to the valid predicted trajectory; obtaining a preset speed deviation calculation weight value; using the obstacle's real-time driving speed, historical driving speed, real-time acceleration, historical acceleration and the Euclidean distance between the obstacle and the turning point to calculate the driving speed of the obstacle to the turning point; using the obstacle in the The turning speed on the effective predicted trajectory, the driving speed from the obstacle to the turning point and the preset speed deviation calculation weight value are used to calculate the speed deviation confidence corresponding to the effective predicted trajectory; the preset confidence calculation weight value of the obstacle in the previous round of prediction action is obtained; multiple similarities between the effective predicted trajectory and multiple historical predicted trajectories are obtained; among the multiple historical predicted trajectories, the target historical predicted trajectory with the highest similarity is determined; the continuous prediction confidence of the effective predicted trajectory is calculated using the prediction confidence of the effective predicted trajectory, the historical trajectory confidence of the target historical predicted trajectory and the preset confidence calculation weight value; the heading angle deviation confidence, the speed deviation confidence and the continuous prediction confidence are added to obtain the trajectory confidence of the effective predicted trajectory.

[0119] In this embodiment, for any valid predicted trajectory, a preset heading angle deviation calculation weight value is obtained. At the same time, on the effective predicted trajectory, the target heading angle of the target position corresponding to the preset time is obtained. By collecting the target heading angle of the predicted point (target position) on the effective predicted trajectory after a period of travel, it helps to understand the direction that the effective predicted trajectory will face at a certain moment in the future, and predict whether there will be a need for turning or steering due to obstacles on the future driving path.

[0120] Furthermore, the weight value is calculated using the obstacle's real-time heading angle, historical heading angle, target heading angle, and preset heading angle deviation to calculate the heading angle deviation confidence of the effective predicted trajectory:

[0121]

[0122] Among them, the above C HI is the heading angle deviation confidence of the Ith valid predicted trajectory; the above H c is the real-time heading angle of the obstacle; the above H o is the historical heading angle of the obstacle; the above H ti is the target heading angle at the target position at the preset time t on the i-th valid prediction trajectory; the above H weight Calculates weight value for preset heading angle deviation.

[0123] Furthermore, the preset speed deviation calculation weight value is obtained. The speed at which the obstacle reaches the turning point is calculated using the obstacle's real-time speed, historical speed, real-time acceleration, historical acceleration, and the Euclidean distance between the obstacle and the turning point:

[0124]

[0125] Among them, the above V r is the speed at which the obstacle reaches the turning point; the above V c is the real-time speed of the obstacle; the above V o is the historical speed of the obstacle; the above A c is the real-time acceleration of the obstacle; the above A o is the historical acceleration of the obstacle; the above D c is the Euclidean distance.

[0126] Then, the speed deviation confidence of the effective predicted trajectory is calculated by using the turning speed of the obstacle on the effective predicted trajectory, the driving speed of the obstacle at the turning point, and the preset speed deviation to calculate the weight value:

[0127] C VI =(V curveI -V r )×V weight ;

[0128] Among them, the above C VI is the speed deviation confidence of the Ith valid predicted trajectory; the above V weight Calculates weight value for preset speed deviation.

[0129] Furthermore, the trajectory prediction action for obstacles near the autonomous driving vehicle is continuous within a set time range. For example, the obstacle vehicle near the autonomous driving vehicle is monitored every 100 milliseconds and its moving trajectory is predicted. In order to enhance the credibility and reliability of the prediction results, the confidence of the two prediction actions can be calculated to verify the continuity of the predicted trajectory in time. If the confidence between the result of the previous round of prediction action and the prediction trajectory of this round is high, it means that the continuity and evolution trend of the obstacle behavior can be better captured. Therefore, among the multiple historical prediction trajectories in the previous round of prediction actions of the obstacle, the target historical prediction trajectory with the highest similarity to the effective prediction trajectory is determined, and the historical trajectory confidence of the target historical prediction trajectory is obtained. The prediction confidence of the effective prediction trajectory, the historical trajectory confidence of the target historical prediction trajectory and the preset confidence calculation weight value are used to calculate the continuous prediction confidence of the effective prediction trajectory:

[0130] C OI =C pi ×(1-Oweight )-C oi ×O weight ;

[0131] Among them, the above C OI is the continuous prediction confidence of the Ith valid prediction trajectory; the above C pi is the prediction confidence of the i-th valid prediction trajectory; weight The weight value for calculating the preset reliability; the above C oi is the historical trajectory confidence of the target historical prediction trajectory corresponding to the i-th valid prediction trajectory.

[0132] Optionally, in order to improve the accuracy of continuous prediction confidence, a similarity threshold can be set. If the multiple similarities between the effective prediction trajectory and multiple historical prediction trajectories are all less than the similarity threshold, it means that there is no historical prediction trajectory matching the effective prediction trajectory in the previous round of prediction action. At this time, C oi is 0.

[0133] Finally, the heading angle deviation confidence, velocity deviation confidence, and continuous prediction confidence of the effective predicted trajectory are added together to obtain the trajectory confidence C of the effective predicted trajectory. I :

[0134] C I =C HI +C VI +C OI .

[0135] Through the above method, by calculating the confidence of the heading angle deviation, speed deviation and the continuity between the predicted actions of the front and rear rounds respectively, the obstacle prediction trajectory can be evaluated more comprehensively to ensure the accuracy and reliability of the optimal predicted trajectory finally selected.

[0136] S208 . Determine, among multiple valid predicted trajectories, a valid predicted trajectory corresponding to a maximum trajectory confidence level as an optimal predicted trajectory of the obstacle.

[0137] This step and Figure 1 The method of step S104 shown is the same and will not be repeated here.

[0138] Optionally, after determining the optimal predicted trajectory of the obstacle in this round of prediction action, the real-time driving data and prediction data of the obstacle in this round of prediction action are summarized, and the summarized data is stored in a database for storing obstacle prediction action data to provide basic data for the next round of prediction action. At the same time, all the data of this round of prediction action are used to optimize the preset trajectory prediction model so that the model parameters better fit the actual situation. Improve the accuracy and stability of model prediction.

[0139] In one embodiment of the present application, optionally, the trajectory prediction method of the embodiment of the present application includes the following four steps:

[0140] S1: In the high-precision map, multiple target lanes are determined according to the location of the moving vehicle.

[0141] Specifically, a search distance threshold R is pre-set, and lane data of multiple target lanes in a high-precision map within a target circular area with a radius of R and a location of the autonomous driving vehicle as the origin is obtained. The lane data includes the front and rear order of multiple lanes, left and right adjacent relationships, and lane attributes, such as lane length, width, curvature value, direction, etc.

[0142] S2: Project the detected obstacles onto a high-precision map.

[0143] Specifically, sensors are used to monitor obstacles on multiple target lanes, and the UTM (UNIVERSAL TRANSVERSE MERCARTOR GRID SYSTEM) coordinate system is used to project one or more monitored obstacles onto a high-precision map.

[0144] S3: Obtain the real-time driving data of the obstacle, the historical driving data of the obstacle in the previous round of predicted actions, and the historical prediction data; use the preset trajectory prediction model to obtain multiple predicted trajectories.

[0145] Specifically, the first position coordinates of the obstacle, the type of obstacle (such as motor vehicle, non-motor vehicle, etc.), real-time heading angle, real-time driving speed, real-time acceleration, etc. are obtained. At the same time, according to the identity of the obstacle, the historical driving data and historical prediction data of the obstacle in the previous round of prediction actions are retrieved from multiple stored data.

[0146] Afterwards, the real-time driving data, historical driving data and historical prediction data of the obstacle are used as inputs of the preset trajectory prediction model to output multiple predicted trajectories. Figure 5 As shown, it is a schematic diagram of multiple predicted trajectories in a high-precision map. The preset trajectory prediction model outputs three predicted trajectories based on the real-time driving data, historical driving data and historical prediction data of the obstacle: predicted trajectory 1, predicted trajectory 2 and predicted trajectory 3.

[0147] S4: Obtain the validity and trajectory confidence of each predicted trajectory; determine the optimal predicted trajectory according to the validity and trajectory confidence of each predicted trajectory, and store the real-time driving data and predicted trajectory data of the obstacle.

[0148] Specifically, obtain the real-time speed V of the obstacle c , real-time acceleration A c , real-time heading angle Hc .

[0149] Get the first position coordinate P of the obstacle in the UTM coordinate system c (X c , Y c ), and the second position coordinate P of the turning point closest to the obstacle r (X r , Y r ).

[0150] Get the road topology within the radius R of the driving vehicle: target lane L r1 , L r2 ,......,L rR , and the lane curvature value K corresponding to each target lane r1 , K r2 ,......,K rR .

[0151] Obtain the historical driving data of the obstacle in the previous round of prediction actions: historical driving speed V o 、Historical acceleration A o 、Historical heading angle H o .

[0152] Obtain the historical prediction data of the obstacle in the previous round of prediction actions: multiple historical prediction trajectories T o1 、T o2 ,......,T oN , the trajectory confidence C of each historical prediction trajectory o1 , C o2 ,......,C oN , and the intention type of each historical predicted trajectory: such as going straight, turning left, turning right, U-turn, etc.

[0153] Using the preset trajectory prediction model, obtain multiple predicted trajectories T of the obstacle in this round of prediction action c1 、T c2 ,......,T cN , the prediction confidence C corresponding to each predicted trajectory p1 , C p2 ,......,C pN , the heading angle H of the target position of each predicted trajectory at the preset time t t1 , H t2 ,......,H tN , and the intent type of each predicted trajectory. Figure 3As shown in FIG. 1 , it is one of the schematic diagrams of the structure of an obstacle on a high-precision map, where the first position coordinates Pc (Xc, Yc) of the obstacle in the UTM coordinate system; the second position coordinates Pr (Xr, Yr) of the turning point closest to the obstacle; the Euclidean distance D between the obstacle and the turning point c Using the preset trajectory prediction model, the three predicted trajectories of the output obstacles are located at T c1 、T c2 and T c3 Among them, T c1 The intention type is straight, T c2 The intention type is turn, T c3 The intent type is U-turn.

[0154] For any predicted trajectory, check the validity of the trajectory:

[0155] Set the maximum centripetal acceleration of the obstacle turning on the predicted trajectory to A curve , lane curvature K ri (i is 1, 2...N); the maximum longitudinal deceleration of the obstacle before turning is A dec ;

[0156] Calculate the turning speed of an obstacle:

[0157]

[0158] If V curveI Greater than V c , then V curveI Equal to V c ;

[0159] Calculate the minimum deceleration distance for obstacles:

[0160]

[0161] Calculate the Euclidean distance between the obstacle and the nearest turning point:

[0162]

[0163] For turning around obstacles, the speed should be less than V curveI , that is, the obstacle should be decelerated to a value less than or equal to V when it reaches the turning point curveI Therefore, if D c <D decI和 / or A c >A dec , then the predicted trajectory is considered to be an invalid predicted trajectory.

[0164] Furthermore, for any valid predicted trajectory, the trajectory confidence of the valid predicted trajectory is obtained:

[0165] Set the heading angle deviation calculation weight value H weight , effectively predict the trajectory at the predicted point heading angle H at the preset time t ti (i is 1, 2...N); Set the speed deviation calculation weight value V weight ; Set the confidence calculation weight value corresponding to the previous prediction action O weight , the confidence C of the historical prediction trajectory in the previous prediction action oi (i takes 1, 2, ... N), the confidence C of the effective prediction trajectory in the current prediction action pi (i takes 1, 2...N);

[0166] Calculate the heading angle deviation confidence of the valid predicted trajectory:

[0167] C HI =-((H c +H o ) / 2-H ti ) 2 ×H weight , (I takes 1, 2...N);

[0168] Calculate the speed from the obstacle to the turning point:

[0169]

[0170] Calculate the velocity deviation confidence of the effective predicted trajectory:

[0171] C VI =(V curveI -V r )×V weight , (I takes 1, 2...N);

[0172] Calculate the continuous prediction confidence of the effective prediction trajectory before and after the two consecutive prediction actions:

[0173] C OI =C pi ×(1-O weight )-C oi ×O weight , (I takes 1, 2...N);

[0174] Compute trajectory confidence for valid predicted trajectories:

[0175] C I =C HI +C VI +C OI , (I takes 1, 2...N);

[0176] The trajectory confidence C of the current multiple valid predicted trajectories ISort and select the valid predicted trajectory with the largest trajectory confidence as the optimal predicted trajectory of the obstacle.

[0177] In actual application scenarios, the maximum centripetal acceleration A of the obstacle at the turn is set curve =2m / s 2 ; The maximum longitudinal deceleration of the obstacle before turning A dec =-2m / s 2 ; Set the heading angle deviation calculation weight value H weight =0.1; Set the speed deviation calculation weight value V weight =0.01; Set the confidence calculation weight value corresponding to the previous round of prediction action O weight = 0.3; Set the curvature value K of lane 101 r1 =1.0, the curvature value K of lane 106 r2 = 0.14, the curvature value K of lane 111 r3 =0.57.

[0178] like Figure 4 As shown in the figure, it is the second schematic diagram of the predicted trajectory of the obstacle in the high-precision map, obtaining the real-time driving data of the obstacle: the real-time driving speed V c =2.80m / s, real-time acceleration A c =-0.1m / s 2 , real-time heading H c =-1.2741rad, the first position coordinate of the obstacle in the UTM coordinate system P c (286201.97, 3472488.68), the second position coordinate P of the obstacle closest to the turning point on the driving direction r (286201.34, 3472487.83).

[0179] Obtain the historical driving data of the obstacle in the previous round of prediction actions: historical driving speed V o =2.8m / s, historical acceleration A o =-0.1m / s 2 , historical heading angle H o =-1.2793rad;

[0180] Obtain the historical prediction data of the obstacle in the previous round of prediction actions: historical prediction trajectory T o1 、T o2 、T o3 , and the corresponding historical trajectory confidence C o1 =0.76, C o2 =0.83, C o3 =0.22.

[0181] Using the preset trajectory prediction model, multiple predicted trajectories T are obtained c1 、T c2 、T c3 , and the corresponding prediction confidence C p1 =0.65, C p2 =0.81, C p3 =0.1, where T c1 Corresponding lanes 101, T c2 Corresponding lanes 106, T c3 Corresponding lane 111.

[0182] The turning speed of the obstacle in each predicted trajectory is calculated as: T c1 The turning speed V curve1 =2.8m / s, T c2 The turning speed V curve2 =2.8m / s, T c3 The turning speed V curve3 =1.87m / s.

[0183] The minimum deceleration distance of the obstacle in each predicted trajectory is calculated as follows: dec1 =0m, T c2 The minimum deceleration distance D dec2 =0m, T c3 The minimum deceleration distance D dec3 =1.08m.

[0184] The Euclidean distance Dc between the obstacle position and the turning point is calculated as 1.058 m.

[0185] The candidate prediction trajectory T is thus determined c1 、T c2 To effectively predict the trajectory, T c3 For invalid prediction trajectories, determine T c3 The trajectory confidence value is 0.

[0186] Get the preset time t = 1.0s, the corresponding heading angles of the target position are: H t1 =-1.3090rad, H t2 =-1.3566rad, the heading angle deviation confidence of each valid predicted trajectory is calculated as: C H1 =-1.0426e-4, C H2 =-4.029e-05.

[0187] Calculate the speed V of the vehicle when the obstacle reaches the turning point r =2.76m / s, the speed deviation confidence of each effective prediction trajectory is calculated as: C V1 =3.8045e-4, CV2 =3.8045e-4.

[0188] The continuous prediction confidence of each valid prediction trajectory is calculated as: C O1 =0.843, C O2 =0.81;

[0189] Calculate T c1 The trajectory confidence is: C1 = 0.683, T c2 The trajectory confidence is: C2=0.816.

[0190] Sorting the trajectory confidence of all current valid trajectories, we get C2>C1, so the candidate predicted trajectory T c2 It will be used as the optimal predicted trajectory of the obstacle in this round of predicted actions.

[0191] In practical application scenarios, the trajectory prediction method of the present application is applicable to trajectory prediction systems, such as Figure 5 As shown, it is a structural diagram of the trajectory prediction system. Specifically, the trajectory prediction system includes an input unit, a processing unit, a storage unit and an output unit. Among them, the input unit is used to obtain the road topology data of multiple target lanes through a high-precision map engine according to the position of the driving vehicle, as well as the real-time driving data of obstacles perceived by the fusion of sensors such as lidar, camera, millimeter-wave radar (such as obstacle position, type, heading, driving speed, acceleration, etc.). The processing unit is used to retrieve the historical driving data and historical prediction data of the obstacle in the previous round of prediction actions in the storage unit according to the real-time driving data of the input obstacle. Using the trained preset trajectory prediction model, multiple candidate prediction trajectories are obtained, and the validity and confidence of all candidate prediction trajectories are calculated to obtain the optimal prediction trajectory. The storage unit is used to store the obstacle driving data and prediction data in the current prediction action for the calculation of the next prediction action. The output unit outputs the optimal prediction trajectory, thereby helping the driving vehicle to generate a safe, comfortable and efficient driving trajectory.

[0192] Further, as Figure 1 In a specific implementation of the method, the present application embodiment provides a trajectory prediction device 300, such as Figure 6 As shown, the device includes: a first acquisition module 301, a first determination module 302, a second acquisition module 303 and a second determination module 304.

[0193] The first acquisition module 301 is used to acquire the real-time driving data of the obstacle, the historical driving data of the obstacle in the previous round of prediction action, and the historical prediction data;

[0194] A first determination module 302, configured to determine a plurality of valid predicted trajectories of the obstacle on a plurality of target lanes according to the real-time driving data, historical driving data and historical prediction data of the obstacle;

[0195] The second acquisition module 303 is used to obtain the trajectory confidence corresponding to each valid predicted trajectory;

[0196] The second determination module 304 is configured to determine, among multiple valid predicted trajectories, a valid predicted trajectory corresponding to a maximum trajectory confidence as an optimal predicted trajectory of the obstacle.

[0197] In a specific application scenario, the first acquisition module 301 is specifically used to:

[0198] Obtain the location information of the moving vehicle on the high-precision map;

[0199] With the location of the moving vehicle as the origin and the preset distance as the radius, a target circular area is generated on the high-precision map;

[0200] Obtaining multiple curvature values ​​of multiple target lanes within the target circular area;

[0201] Obtaining the identity identification, first position coordinate data and real-time driving data of obstacles on multiple target lanes, wherein the real-time driving data includes: real-time driving speed, real-time acceleration and real-time heading angle;

[0202] According to the identity identifier, the historical driving data and the historical prediction data of the obstacle in the previous round of prediction actions are obtained from the multiple stored data corresponding to the previous round of prediction actions;

[0203] The historical driving data includes the historical driving speed, historical acceleration and historical heading angle of the obstacle in the previous round of prediction actions; the historical prediction data includes multiple historical prediction trajectories of the obstacle in the previous round of prediction actions, and the historical trajectory confidence of each historical prediction trajectory.

[0204] In a specific application scenario, the first determining module 302 is specifically used to:

[0205] The real-time driving data, historical driving data and historical prediction data of the obstacle are input into a preset trajectory prediction model, and multiple predicted trajectories of the obstacle on multiple target lanes and the prediction confidence corresponding to each predicted trajectory are output;

[0206] Get the intent type of each predicted trajectory;

[0207] For any predicted trajectory, if the intention type is a straight-ahead type, the predicted trajectory is determined to be a valid trajectory;

[0208] If the intention type is a turning type or a U-turn type, obtain the minimum deceleration distance of the obstacle in the predicted trajectory, the Euclidean distance between the obstacle and the nearest turning point on the predicted trajectory, and the maximum longitudinal deceleration of the obstacle before turning;

[0209] If the Euclidean distance between the obstacle and the turning point is less than the minimum deceleration distance and / or the real-time acceleration of the obstacle is greater than the maximum longitudinal deceleration of the obstacle before turning, the predicted trajectory is determined to be an invalid predicted trajectory;

[0210] If the Euclidean distance between the obstacle and the turning point is greater than or equal to the minimum deceleration distance, and the real-time acceleration of the obstacle is less than or equal to the maximum longitudinal deceleration of the obstacle before turning, the predicted trajectory is determined to be a valid predicted trajectory.

[0211] In a specific application scenario, the first determining module 302 is further configured to:

[0212] Obtain the second position coordinate of the turning point on the predicted trajectory that is closest to the obstacle;

[0213] Determine the Euclidean distance between the obstacle and the turning point according to the first position coordinate and the second position coordinate;

[0214] Obtain the maximum centripetal acceleration of the obstacle at the turn on the predicted trajectory and the maximum longitudinal deceleration before the turn;

[0215] The turning speed and minimum deceleration distance of the obstacle on the predicted trajectory are determined based on the maximum centripetal acceleration at the turn, the curvature value of the lane where the predicted trajectory is located, the real-time driving speed, and the maximum longitudinal deceleration before the turn.

[0216] In a specific application scenario, the first determining module 302 is further configured to:

[0217] The turning speed of the obstacle on the predicted trajectory is calculated using the turning centripetal acceleration and the curvature value of the lane where the predicted trajectory is located;

[0218] If the turning speed is greater than the real-time speed of the obstacle, the real-time speed is used as the turning speed of the obstacle on the predicted trajectory;

[0219] The minimum deceleration distance of the obstacle on the predicted trajectory is calculated using the obstacle's turning speed, real-time driving speed, and maximum longitudinal deceleration before turning.

[0220] In a specific application scenario, the second acquisition module 303 specifically includes:

[0221] For any valid predicted trajectory, obtain the preset heading angle deviation calculation weight value, and effectively predict the target heading angle of the target position corresponding to the preset time on the trajectory;

[0222] The weight value is calculated using the obstacle's real-time heading angle, historical heading angle, target heading angle and preset heading angle deviation to calculate the heading angle deviation confidence level corresponding to the effective predicted trajectory.

[0223] Get the preset speed deviation calculation weight value;

[0224] The speed at which the obstacle reaches the turning point is calculated using the obstacle's real-time speed, historical speed, real-time acceleration, historical acceleration, and the Euclidean distance between the obstacle and the turning point.

[0225] The speed deviation confidence corresponding to the effective predicted trajectory is calculated by using the turning speed of the obstacle on the effective predicted trajectory, the driving speed of the obstacle at the turning point, and the preset speed deviation to calculate the weight value;

[0226] Obtain the preset confidence calculation weight value of the obstacle in the previous round of prediction action;

[0227] Obtain multiple similarities between the effective prediction trajectory and multiple historical prediction trajectories;

[0228] Determine the target historical prediction trajectory with the highest similarity among multiple historical prediction trajectories;

[0229] The continuous prediction confidence corresponding to the effective prediction trajectory is calculated by using the prediction confidence corresponding to the effective prediction trajectory, the historical trajectory confidence of the target historical prediction trajectory and the preset confidence calculation weight value;

[0230] The heading angle deviation confidence, velocity deviation confidence and continuous prediction confidence are added together to obtain the trajectory confidence of the effective predicted trajectory.

[0231] In specific application scenarios, such as Figure 6 As shown, the device further includes: a judgment module 305 , a prediction module 306 and a third determination module 307 .

[0232] A determination module 305, configured to determine whether historical data of the obstacle is stored in the plurality of stored data according to the identity of the obstacle;

[0233] The prediction module 306 is used to input the real-time driving speed, real-time acceleration and real-time heading angle of the obstacle into a preset trajectory prediction model if the historical data of the obstacle is not stored in the plurality of stored data, and output a plurality of predicted trajectories and a prediction confidence of each predicted trajectory;

[0234] The third determination module 307 is used to determine, among multiple predicted trajectories, a predicted trajectory corresponding to a maximum prediction confidence as an optimal predicted trajectory of the obstacle.

[0235] The trajectory prediction device 300 provided in the embodiment of the present application, after monitoring the obstacle vehicle near the autonomous driving vehicle, collects the real-time driving data of the obstacle, combines the historical driving data and historical trajectory data of the obstacle in the previous round of prediction action, generates multiple predicted trajectories of the obstacle in this round of prediction action, verifies the validity of the multiple predicted trajectories, and obtains multiple valid predicted trajectories. Thereafter, based on the intention type corresponding to each valid predicted trajectory, the trajectory confidence of each valid predicted trajectory is determined, and finally the valid predicted trajectory with the largest trajectory confidence is used as the optimal predicted trajectory of this round of prediction action. Compared with the prediction method of the prior art that only uses the real-time driving data of the obstacle for trajectory prediction, the present application obtains the validity and trajectory confidence of multiple predicted trajectories of the obstacle under multiple intention types in a complex road environment, and then determines the optimal predicted trajectory of the obstacle based on the validity and trajectory confidence of the multiple predicted trajectories, ensures that the predicted moving trajectory conforms to the complex road conditions, and improves the accuracy and reliability of the predicted trajectory.

[0236] In an exemplary embodiment, the present application further provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor is used to execute the program stored in the memory and execute the trajectory prediction method in the above embodiment.

[0237] In an exemplary embodiment, the present application further provides a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the trajectory prediction method are implemented.

[0238] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by hardware, or by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile readable storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for enabling an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0239] Those skilled in the art will appreciate that the accompanying drawings are merely schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required for implementing the present application.

[0240] Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices in the implementation scenario according to the implementation scenario description, or can be changed accordingly and located in one or more devices different from the implementation scenario. The modules in the above implementation scenario can be combined into one module, or can be further split into multiple units.

[0241] The above application serial numbers are for description only and do not represent the advantages or disadvantages of the implementation scenarios.

[0242] The above disclosure only discloses several specific implementation scenarios of the present application. However, the present application is not limited thereto, and any changes that can be conceived by technicians in this field should fall within the scope of protection of the present application.

Claims

1. A trajectory prediction method, characterized in that: include: Acquire real-time driving data of the obstacle, historical driving data of the obstacle in the previous round of prediction actions, and historical prediction data; Determining a plurality of valid predicted trajectories of the obstacle on a plurality of target lanes according to the real-time driving data, the historical driving data and the historical prediction data of the obstacle; Get the trajectory confidence corresponding to each valid predicted trajectory; Among the multiple valid predicted trajectories, determining a valid predicted trajectory corresponding to a maximum trajectory confidence as the optimal predicted trajectory of the obstacle; The step of obtaining the trajectory confidence corresponding to each valid predicted trajectory specifically includes: For any valid predicted trajectory, obtain the preset heading angle deviation calculation weight value, and effectively predict the target heading angle of the target position corresponding to the preset time on the trajectory; The weight value is calculated using the real-time heading angle, historical heading angle, target heading angle and preset heading angle deviation of the obstacle to calculate the heading angle deviation confidence corresponding to the effective predicted trajectory; Get the preset speed deviation calculation weight value; The speed at which the obstacle reaches the turning point is calculated using the obstacle's real-time speed, historical speed, real-time acceleration, historical acceleration, and the Euclidean distance between the obstacle and the turning point. The speed deviation confidence corresponding to the effective predicted trajectory is calculated by using the turning speed of the obstacle on the effective predicted trajectory, the driving speed of the obstacle at the turning point, and the preset speed deviation to calculate the weight value; Obtain the preset confidence calculation weight value of the obstacle in the previous round of prediction action; Obtain multiple similarities between the effective prediction trajectory and multiple historical prediction trajectories; Determine the target historical prediction trajectory with the highest similarity among the multiple historical prediction trajectories; The continuous prediction confidence corresponding to the effective prediction trajectory is calculated by using the prediction confidence corresponding to the effective prediction trajectory, the historical trajectory confidence of the target historical prediction trajectory and the preset confidence calculation weight value; The heading angle deviation confidence, the speed deviation confidence and the continuous prediction confidence are added to obtain the trajectory confidence of the effective predicted trajectory.

2. The method according to claim 1, characterized in that The step of obtaining the real-time driving data of the obstacle, the historical driving data of the obstacle in the previous round of prediction action, and the historical prediction data specifically includes: Obtain the location information of the moving vehicle on the high-precision map; Taking the location of the moving vehicle as the origin and the preset distance as the radius, a target circular area is generated on the high-precision map; Acquire multiple curvature values ​​of multiple target lanes within the target circular area; Obtaining the identity identification, first position coordinate data and real-time driving data of the obstacles on the multiple target lanes, wherein the real-time driving data includes: real-time driving speed, real-time acceleration and real-time heading angle; According to the identity identifier, obtaining historical driving data and historical prediction data of the obstacle in the previous round of prediction actions from a plurality of stored data corresponding to the previous round of prediction actions; The historical driving data includes the historical driving speed, historical acceleration and historical heading angle of the obstacle in the previous round of prediction actions; the historical prediction data includes multiple historical prediction trajectories of the obstacle in the previous round of prediction actions, and the historical trajectory confidence of each historical prediction trajectory.

3. The method according to claim 2, characterized in that The step of determining a plurality of valid predicted trajectories of the obstacle on the plurality of target lanes according to the real-time driving data, the historical driving data and the historical prediction data of the obstacle specifically comprises: Inputting the real-time driving data, the historical driving data and the historical prediction data of the obstacle into a preset trajectory prediction model, and outputting a plurality of predicted trajectories of the obstacle on the plurality of target lanes, and a prediction confidence corresponding to each predicted trajectory; Obtaining the intent type of each predicted trajectory; For any predicted trajectory, if the intention type is a straight-ahead type, determining the predicted trajectory as a valid trajectory; If the intention type is a turning type or a U-turn type, obtaining the minimum deceleration distance of the obstacle in the predicted trajectory, the Euclidean distance between the obstacle and the nearest turning point on the predicted trajectory, and the maximum longitudinal deceleration of the obstacle before turning; If the Euclidean distance between the obstacle and the turning point is less than the minimum deceleration distance and / or the real-time acceleration of the obstacle is greater than the maximum longitudinal deceleration of the obstacle before turning, determining that the predicted trajectory is an invalid predicted trajectory; If the Euclidean distance between the obstacle and the turning point is greater than or equal to the minimum deceleration distance, and the real-time acceleration of the obstacle is less than or equal to the maximum longitudinal deceleration of the obstacle before turning, the predicted trajectory is determined to be a valid predicted trajectory.

4. The method according to claim 3, characterized in that The step of obtaining the minimum deceleration distance of the obstacle in the predicted trajectory, the Euclidean distance between the obstacle and the nearest turning point on the predicted trajectory, and the maximum longitudinal deceleration of the obstacle before turning specifically includes: Obtain the second position coordinate of the turning point on the predicted trajectory that is closest to the obstacle; Determine the Euclidean distance between the obstacle and the turning point according to the first position coordinates and the second position coordinates; Obtaining a maximum centripetal acceleration of the obstacle at a turn on the predicted trajectory and a maximum longitudinal deceleration before the turn; The turning speed and minimum deceleration distance of the obstacle on the predicted trajectory are determined according to the maximum centripetal acceleration at the turning point, the curvature value of the lane where the predicted trajectory is located, the real-time driving speed, and the maximum longitudinal deceleration before the turning point.

5. The method according to claim 4, characterized in that The step of determining the turning speed and the minimum deceleration distance of the obstacle on the predicted trajectory according to the maximum centripetal acceleration at the turning point, the curvature value of the lane where the predicted trajectory is located, the real-time driving speed, and the maximum longitudinal deceleration before the turning specifically includes: Calculating the turning speed of the obstacle on the predicted trajectory by using the maximum centripetal acceleration at the turning point and the curvature value of the lane where the predicted trajectory is located; If the turning speed is greater than the real-time driving speed of the obstacle, the real-time driving speed is used as the turning speed of the obstacle on the predicted trajectory; The minimum deceleration distance of the obstacle on the predicted trajectory is calculated using the turning speed of the obstacle, the real-time driving speed, and the maximum longitudinal deceleration before turning.

6. The method according to claim 2, characterized in that Before obtaining the historical driving data and historical prediction data of the obstacle in the previous round of prediction actions from the plurality of stored data corresponding to the previous round of prediction actions according to the identity identifier, the method further includes: According to the identity identifier of the obstacle, determining whether historical data of the obstacle is stored in the plurality of stored data; If the historical data of the obstacle is not stored in the plurality of stored data, the real-time speed, the real-time acceleration and the real-time heading angle of the obstacle are input into a preset trajectory prediction model, and a plurality of predicted trajectories and a prediction confidence of each predicted trajectory are output; Among multiple predicted trajectories, a predicted trajectory corresponding to a maximum prediction confidence is determined as an optimal predicted trajectory of the obstacle.

7. A trajectory prediction device, characterized in that: include: A first acquisition module is used to acquire real-time driving data of an obstacle, historical driving data of the obstacle in a previous round of prediction actions, and historical prediction data; A first determination module, configured to determine a plurality of valid predicted trajectories of the obstacle on a plurality of target lanes according to the real-time driving data, the historical driving data and the historical prediction data of the obstacle; The second acquisition module is used to obtain the trajectory confidence corresponding to each valid predicted trajectory; A second determination module is used to determine, among the multiple valid predicted trajectories, a valid predicted trajectory corresponding to a maximum trajectory confidence as an optimal predicted trajectory of the obstacle; The second acquisition module is specifically used for: For any valid predicted trajectory, obtain the preset heading angle deviation calculation weight value, and effectively predict the target heading angle of the target position corresponding to the preset time on the trajectory; The weight value is calculated using the obstacle's real-time heading angle, historical heading angle, target heading angle and preset heading angle deviation to calculate the heading angle deviation confidence level corresponding to the effective predicted trajectory. Get the preset speed deviation calculation weight value; The speed at which the obstacle reaches the turning point is calculated using the obstacle's real-time speed, historical speed, real-time acceleration, historical acceleration, and the Euclidean distance between the obstacle and the turning point. The speed deviation confidence corresponding to the effective predicted trajectory is calculated by using the turning speed of the obstacle on the effective predicted trajectory, the driving speed of the obstacle when it reaches the turning point, and the preset speed deviation to calculate the weight value; Obtain the preset confidence calculation weight value of the obstacle in the previous round of prediction action; Obtain multiple similarities between the effective prediction trajectory and multiple historical prediction trajectories; Determine the target historical prediction trajectory with the highest similarity among multiple historical prediction trajectories; The continuous prediction confidence corresponding to the effective prediction trajectory is calculated by using the prediction confidence corresponding to the effective prediction trajectory, the historical trajectory confidence of the target historical prediction trajectory and the preset confidence calculation weight value; The heading angle deviation confidence, velocity deviation confidence and continuous prediction confidence are added together to obtain the trajectory confidence of the effective predicted trajectory.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When a processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Citation Information

Patent Citations

  • Trajectory prediction method and device

    CN112839855A

  • Automatic driving track prediction method and device, equipment and storage medium

    CN114771572A

Cited By

  • Trajectory planning method and apparatus, electronic device, and autonomous vehicle

    EP4744995A3