Search

CN-116654018-B - Target track prediction method, computer equipment, machine-readable storage medium and motor vehicle

CN116654018BCN 116654018 BCN116654018 BCN 116654018BCN-116654018-B

Abstract

The invention discloses a target track prediction method, computer equipment, a machine-readable storage medium and a motor vehicle, and relates to the technical field of automatic driving, comprising the steps of modeling and generating an interactive heat map; and a track prediction step of constructing a driving force constraint and an environmental force constraint and predicting track coordinates of the next time step to form a track of the predicted target. The track prediction method and the track prediction device greatly improve the accuracy of track prediction.

Inventors

  • WANG ZIHAO

Assignees

  • 浙江零跑科技股份有限公司

Dates

Publication Date
20260508
Application Date
20230630

Claims (10)

  1. 1. A target track prediction method is used for predicting the track of a target around a vehicle when the vehicle runs and drives automatically in an urban intersection scene, and is characterized in that the target track prediction method comprises the following steps of, Modeling, namely acquiring a track heat map according to a predicted target and a satellite map of an urban intersection scene, performing semantic segmentation on the satellite map of the urban intersection scene according to the types of obstacles in the urban intersection scene to acquire a scene semantic map, constructing a distance heat map according to the influence of other targets at different positions around the predicted target on the predicted target, screening the other targets in the distance heat map, and generating an interactive heat map based on the screened other targets; a position prediction step, namely generating an end point heat map according to the track heat map, the scene semantic map and the interaction heat map of the predicted target, and determining the predicted position of the predicted target by the end point heat map; a track prediction step, comprising: calculating the direction and the magnitude of the expected speed of the predicted target moving to the predicted position, and constructing a driving force constraint according to the direction and the magnitude of the expected speed; constructing environmental force constraint according to the scene semantic graph; And establishing a dynamic model in a discrete state according to the driving force constraint and the environmental force constraint, and predicting the track coordinate of the next time step based on the dynamic model in the discrete state to form the track of the predicted target.
  2. 2. The target trajectory prediction method according to claim 1, wherein the direction and magnitude of the desired speed of the predicted target movement to the predicted position are calculated according to the following formula: Wherein, the As a vector, representing the direction of the desired velocity, Is a scalar, representing the magnitude of the desired speed, t is the current time step, For the time step of the future time step, Is the observation track of the current time step, the observation track is a two-dimensional coordinate, The predicted position obtained by the position predicting step is a two-dimensional coordinate.
  3. 3. The target trajectory prediction method according to claim 2, characterized in that the driving force constraint is constructed according to the direction and magnitude of the desired speed by the following formula: Wherein, the Is the driving force constraint of the step at t, Is a vector, representing the desired speed, Is a vector, representing the speed of the current time step, Network parameters for constructing driving force constraints; For relaxation time, the current movement speed is regulated for the prediction target to reach again Is a time interval of (2); is a neural network.
  4. 4. A method of target trajectory prediction as claimed in claim 3, wherein the environmental force constraints are constructed from the phase scene semantic graph by the formula: Wherein, the Is the environmental force constraint of the t time step, k is a natural number, To predict the location of a stationary obstacle around the target, Cut out from the scene semantic graph for t time steps An image of a size of the image is displayed, A network is extracted for the image features, To predict the environmental force coefficients of stationary obstacles around the target, Network parameters of the network are extracted for the image features, and C is the category number of the obstacle.
  5. 5. The target trajectory prediction method according to claim 4, wherein the dynamic model in the discrete state is established according to the driving force constraint and the environmental force constraint by the following formula: Wherein, the For the driving force constraint, For the purpose of environmental force constraints, Is a time interval.
  6. 6. The method according to claim 5, wherein the track coordinates of the next time step are predicted based on a dynamics model in a discrete state by the following formula: Wherein, the For the track coordinates of the next time step, Is the track coordinate of the current time step.
  7. 7. The method of claim 4, further comprising a multi-stage training step employing teacherforcing training strategies comprising training the position prediction step, training a neural network Training of (c) and to neural networks And image feature extraction network And (5) training together.
  8. 8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the target trajectory prediction method of any one of claims 1 to 7 when executing the computer program.
  9. 9. A computer readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements the target trajectory prediction method of any one of claims 1 to 7.
  10. 10. A motor vehicle having an autopilot function, wherein the motor vehicle predicts a trajectory of a target around the vehicle in an urban intersection scene by the target trajectory prediction method according to any one of claims 1 to 7 when the autopilot function is operated.

Description

Target track prediction method, computer equipment, machine-readable storage medium and motor vehicle Technical Field The invention relates to the technical field of automatic driving, in particular to a target track prediction method, computer equipment, a machine-readable storage medium and a motor vehicle. Background In the field of automatic driving, predicting the movement trend of a target around a vehicle is one of important links for supporting the operation of an automatic driving system. In urban intersection traffic scenes, when the optimal track planning is performed based on the self state, not only the motion track of surrounding vehicles is needed to be considered, but also the motion trends of pedestrians and non-motor vehicles are needed to be judged, so that collision is effectively avoided. The prior art converts the problem of predicting the trajectory of a behavioural target into a time series prediction problem of using the past motion of the predicted target to infer the location of its future trajectory point. Because of the randomness and uncertainty in the motion of the predicted target, the predicted trajectory is predicting all possible future trajectory transitions from a single predicted future trajectory that is closest to the true value to a multi-modal. However, such an end-guided trajectory prediction method relies on the accuracy of the predicted end point, and the deviation between the predicted end point and the true trajectory end point will have a great influence on the prediction accuracy of the complete trajectory. In the prior art, the behavior prediction method based on the kinematic model comprises excessive model complexity, and cannot fit a large amount of track data, so that the prediction accuracy is low. Moreover, most social interaction-based behavior prediction methods do not sufficiently consider environmental information other than interaction, and ignore effective information such as road structures, which may cause predicted trajectories of targets around the vehicle itself to deviate from a lane or a passable area. Disclosure of Invention The present invention aims to solve one of the technical problems in the related art to a certain extent. Therefore, the target track prediction method provided by the invention improves the accuracy of track prediction. In order to achieve the above purpose, the invention adopts the following technical scheme: a target track prediction method for predicting track of a target around a vehicle when the vehicle is automatically driven in an urban intersection scene, the target track prediction method comprising, Modeling, namely acquiring a track heat map according to a predicted target and a satellite map of an urban intersection scene, performing semantic segmentation on the satellite map of the urban intersection scene according to the types of obstacles in the urban intersection scene to acquire a scene semantic map, constructing a distance heat map according to the influence of other targets at different positions around the predicted target on the predicted target, screening the other targets in the distance heat map, and generating an interactive heat map based on the screened other targets; A position prediction step, namely generating an end point heat map according to a track heat map, a scene semantic map and an interactive heat map of a predicted target, and determining the predicted position of the predicted target by the end point heat map; a track prediction step, comprising: calculating the direction and the magnitude of the expected speed of the predicted target moving to the predicted position, and constructing a driving force constraint according to the direction and the magnitude of the expected speed; constructing environmental force constraint according to the scene semantic graph; And establishing a dynamic model in a discrete state according to the driving force constraint and the environmental force constraint, and predicting the track coordinate of the next time step based on the dynamic model in the discrete state to form the track of the predicted target. Optionally, the direction and magnitude of the desired speed of movement of the predicted target to the predicted position is calculated according to the following formula: Wherein, the Is a scalar, representing the direction of the desired velocity,Is a scalar, represents the magnitude of the desired speed, T fut is the future time step, X t is the observed trace of the current time step,Is the predicted position obtained by the position predicting step. Alternatively, the driving force constraint is constructed according to the direction and magnitude of the desired speed by the following formula: Wherein, the Is the driving force constraint of the step at t,Is a vector, representing the desired speed,Is vector, representing the current time step speed, W τ is the network parameter for constructing driving force constraint, tau is relaxa