Heterogeneous graph aggregation network-based high-density composite scene trajectory prediction method

Abstract

The invention discloses a high-density composite scene trajectory prediction method based on a heterogeneous graph aggregation network. The method comprises the steps: constructing the heterogeneous graph aggregation network which comprises three sub-structures: a self-adaptive neighbor selector, an encoder and a decoder; firstly, automatically selecting a neighbor of a target object through the self-adaptive neighbor selector, and generating a structure of a heterogeneous graph; clustering the heterogeneous feature information between different types of neighbors by using a two-stage aggregator through an encoder, and performing, by the decocder, decoding by using a historical information residual connection technology based on LSTM; and acquiring the future two-dimensional coordinate time sequence prediction information output of a target object by using input historical track time sequence information features. According to the method, the accuracy of high-density composite scene trajectory prediction can be remarkably improved, and high-precision trajectory prediction of multi-class objects in a complex traffic scene is realized.

Description

technical field [0001] The invention relates to the technical field of trajectory prediction, in particular to a high-density compound scene trajectory prediction method based on a heterogeneous graph aggregation network. Background technique [0002] With the rapid development of autonomous driving technology, a considerable part of motor vehicles will be replaced by autonomous vehicles in the future. In the field of autonomous driving, trajectory prediction is widely used as a core technology in the navigation, control and decision-making of autonomous vehicles. Self-driving vehicles control their own behavior more precisely by predicting the future trajectories of vehicles around them, thereby avoiding safety hazards such as traffic accidents. [0003] Therefore, autonomous vehicles need to use efficient and accurate trajectory prediction technology to make behavior decisions. [0004] Existing trajectory prediction methods are mainly divided into single-scene predictio...

AI Technical Summary

Problems solved by technology

[0008] In view of the above technical problems, the present invention provides a high-density composite scene trajectory prediction method based on a heterogeneous graph aggregation network to solve the weight sharing problem caused by a single neural network and the problem of multiple selection or missing selection of neighbors in a fixed area

Images