Search

CN-122009165-A - Track prediction method, electronic device, vehicle, medium, and program product

CN122009165ACN 122009165 ACN122009165 ACN 122009165ACN-122009165-A

Abstract

The application relates to the technical field of vehicle driving, in particular to a track prediction method, electronic equipment, a vehicle, a medium and a program product, wherein in the method, a first predicted track of a moving obstacle and a target environment image shot by a target vehicle are obtained, the first predicted track is determined based on a BEV feature map of an environment where the moving obstacle is positioned, the first predicted track comprises a plurality of track points, the target environment image comprises a front view image, and the front view image comprises a local area corresponding to at least one track point in the plurality of track points; determining local image features corresponding to local areas of each track point in the target environment image; based on the local image features, a second predicted trajectory of the moving obstacle is predicted. It is known that the local image features include environmental features corresponding to areas where moving obstacles may reach in the future, and the vehicle can expand the perception of the remote traffic elements based on the local image features, so as to predict a more accurate trajectory.

Inventors

  • CHEN SIEN

Assignees

  • 智驾新程(上海)智能科技有限公司

Dates

Publication Date
20260512
Application Date
20260409

Claims (11)

  1. 1. A method of trajectory prediction, the method comprising: acquiring a first predicted track of a moving obstacle and a target environment image shot by a target vehicle, wherein the first predicted track is determined based on a BEV feature map of an environment in which the moving obstacle is positioned, the first predicted track comprises a plurality of track points, the target environment image comprises a front view image, and the front view image comprises a local area corresponding to at least one track point in the plurality of track points; Determining local image features corresponding to local areas of each track point in the target environment image; a second predicted trajectory of the moving obstacle is predicted based on the local image features.
  2. 2. The method of claim 1, wherein the first predicted trajectory has trajectory points coordinates in a vehicle coordinate system, and The determining the local image feature corresponding to the local area of each track point in the target environment image comprises the following steps: Extracting image features in the target environment image to obtain a forward-looking feature image; determining the coordinates of the image track points of the track points in the forward-looking characteristic image; determining a local area of the track point from the front view feature map according to the coordinates of the track point of the image and the area range parameter; and obtaining the local image characteristics based on the image characteristics in the local area.
  3. 3. The method of claim 2, wherein the target environment image is captured by a forward-looking camera of the target vehicle, and The determining the image track point coordinates of the track point in the front-view characteristic image comprises the following steps: acquiring coordinates of the track points of the vehicle in a vehicle coordinate system; And converting the coordinates of the vehicle track points according to the camera parameters of the forward-looking camera to obtain the coordinates of the image track points in the image coordinate system of the forward-looking characteristic image.
  4. 4. The trajectory prediction method according to claim 2, wherein the deriving the local image features based on the image features in the local region includes: The local image features are extracted from the local region using bilinear interpolation.
  5. 5. The trajectory prediction method according to claim 1, wherein the acquiring the first predicted trajectory of the moving obstacle and the target environment image taken by the target vehicle includes: acquiring a BEV characteristic map of the environment where the mobile obstacle is located; Matching BEV features related to the historical track of the moving obstacle in the BEV feature map based on the query features of the moving obstacle; And inputting the BEV characteristics related to the historical track of the moving obstacle into a first prediction model to obtain the first prediction track.
  6. 6. The trajectory prediction method according to claim 5, wherein predicting a second predicted trajectory of the moving obstacle based on the local image features includes: and predicting the second predicted track according to the coordinates of the vehicle track points of the track points, the local image characteristics of the track points and BEV characteristics related to the historical track of the moving obstacle.
  7. 7. The trajectory prediction method according to claim 6, wherein predicting the second predicted trajectory based on coordinates of vehicle trajectory points of the trajectory points, local image features of the trajectory points, and BEV features related to a historical trajectory of the moving obstacle, comprises: mapping vehicle track point coordinates of the track points into track point position coding vectors matched with BEV feature dimensions related to the historical track of the moving obstacle; according to the track point position coding vector, fusing the local image characteristics of the track points into BEV characteristics related to the historical track of the moving obstacle to obtain updated BEV characteristics related to the historical track of the moving obstacle; and inputting the updated BEV characteristics related to the historical track of the moving obstacle into a second prediction model to obtain the second prediction track.
  8. 8. An electronic device comprising a memory for storing instructions; at least one processor configured to execute the instructions to cause the electronic device to implement the trajectory prediction method of any one of claims 1-7.
  9. 9. A vehicle includes a memory for storing instructions; At least one processor configured to execute the instructions to cause the vehicle to implement the trajectory prediction method of any one of claims 1-7.
  10. 10. A computer readable storage medium having stored thereon instructions which, when executed on a computer, cause the computer to perform the trajectory prediction method of any one of claims 1 to 7.
  11. 11. A computer program product, characterized in that the computer program product, when run on a device, causes the device to perform the trajectory prediction method of any one of claims 1 to 7.

Description

Track prediction method, electronic device, vehicle, medium, and program product Technical Field The present application relates to the field of vehicle driving technologies, and in particular, to a track prediction method, an electronic device, a vehicle, a medium, and a program product. Background In the driving process of the vehicle, the future motion trail of the vehicle can be predicted based on surrounding dynamic and static object information, and the current vehicle is assisted to determine a driving strategy. In some embodiments, the current vehicle encodes Bird's Eye View (BEV) feature data based on the collected sensor data, where the BEV feature data includes multi-dimensional information of moving objects (vehicles, pedestrians, etc.), and static environments (e.g., lane lines, road boundaries, traffic signs, etc.), to provide a unified representation for trajectory prediction. And, the vehicle may determine characteristics related to the history track (e.g., a vehicle type and a traveling speed, a traveling direction, etc. of the other vehicle) when the other vehicle is driven in the environment based on the BEV feature map, so that the other vehicle track is predicted based on the characteristics of the other vehicle, so that the current vehicle determines a driving strategy according to the predicted other vehicle track. However, the vehicle has limited perception information according to the BEV feature map, and cannot perceive information such as far-end lane information, signboards, traffic lights and the like, so that the predicted track of the other vehicle does not fit with the lane, or the running speed does not accord with the speed limit information. Therefore, the current vehicle is caused to decelerate and avoid untimely or unnecessary deceleration according to the predicted track of the other vehicle, and the riding body feeling and the comfort are affected. Disclosure of Invention The embodiment of the application provides a track prediction method, electronic equipment, a vehicle, a medium and a program product, which are used for avoiding the problem of inaccurate track prediction of a moving obstacle. The embodiment of the application provides a track prediction method, which comprises the steps of obtaining a first predicted track of a moving obstacle and a target environment image shot by a target vehicle, wherein the first predicted track is determined based on a BEV feature map of an environment where the moving obstacle is located, the first predicted track comprises a plurality of track points, the target environment image comprises a forward-looking image, the forward-looking image comprises a local area corresponding to at least one track point in the plurality of track points, determining local image features corresponding to the local area of each track point in the target environment image, and predicting a second predicted track of the moving obstacle based on the local image features. It will be appreciated that the predicted first predicted trajectory is a rough trajectory point. Because the front view image comprises a local area corresponding to at least one track point in the track points, namely the front view image contains more environment information of the moving obstacle, after the predicted first predicted track is obtained, local image characteristics corresponding to the local area of each track point in the target environment image are determined, namely the environment characteristics corresponding to the area where the future moving track of the moving obstacle possibly reaches can be extracted, and the vehicle can expand the perception of the remote traffic elements based on the local image characteristics, so that the moving obstacle is predicted to be a more accurate track based on the extracted local image characteristics. In some possible implementations of the first aspect, the track points of the first predicted track have coordinates of vehicle track points in a vehicle coordinate system, and determining local image features corresponding to local areas of each track point in the target environment image includes extracting image features in the target environment image to obtain a forward-looking feature image, determining coordinates of the track points in the forward-looking feature image, determining local areas of the track points from the forward-looking feature image according to the coordinates of the track points and the area range parameters, and obtaining local image features based on the image features in the local areas. In some possible implementations of the first aspect, the target environment image is captured by a front-view camera of the target vehicle, and determining the image track point coordinates of the track points in the front-view feature image includes obtaining the vehicle track point coordinates of the track points in a vehicle coordinate system, and performing coordinate system conversion on the vehicle tra