Pedestrian trajectory prediction method based on multi-subdivision feature control

Abstract

The invention discloses a pedestrian trajectory prediction method based on multi-subdivision feature control, and the method comprises the following steps: carrying out the calculation based on an observation trajectory sequence, and obtaining a motion trend feature, an interaction feature, an end point implicit feature and a future trajectory prediction end point of each subdivision feature; a multi-subdivision feature fusion module is designed based on the motion trend features of the subdivision features, trajectory prediction trend features are obtained through calculation, and a future trajectory sequence is obtained through prediction based on the prediction trend features; and calculating a prediction error based on the prediction trajectory sequence and the real trajectory sequence, and adding rule constraints of a mutual information error and an end point prediction error to obtain a total error after the rule constraints are added. According to the invention, the features contained in the historical observation sequence are subdivided and expressed, the subdivided features are subjected to progressive feature fusion layer by layer, the training process is constrained based on the model error added with related rules, and finally the pedestrian trajectory prediction method with high prediction precision and high interpretability is formed.

Description

Technical field [0001] The present disclosure relates to the field of computer visual technology and the automatic driving areas, and more particularly to a pedestrian trajectory prediction method based on multi-segment feature control. Background technique [0002] As an important task of automatic driving technology, pedestrian trajectory prediction has become a research hotspot in computer visual fields in recent years. Understanding and predicting the movement behavior of pedestrians is critical to navigation of autonomous systems in the interactive environment. Unmanned vehicles or robots can plan a safe and interactive future path in advance by predicting the future trajectory of pedestrians, and alerting anomalies. [0003] The difficulty of pedestrian trajectory prediction is that the movement of pedestrians has a high random random, so there may be a variety of potential future behaviors. Early work is usually based on the observation track, predicted a future trajectory...

AI Technical Summary

Problems solved by technology

[0007] In order to solve the problems of the prior art, the present invention proposes a pedestrian trajectory prediction method based on multi-subdivision feature control, the purpose of which is to solve the lack of feature extraction from the historical observation trajectory sequence in the prior art, and at the same time, the degree of influence of different features on trajectory prediction Undifferentiated, the trajectory prediction generation process is too single, and the model loss function lacks corresponding rules and constraints

Images