LSTM neural network pedestrian trajectory prediction method based on human-vehicle interaction

Abstract

The invention discloses an LSTM neural network pedestrian trajectory prediction method based on human-vehicle interaction. The method comprises the following steps: constructing an LSTM neural networkof human-vehicle interaction; establishing input of a multi-layer neural network, including inputting a current pedestrian trajectory, inputting human-human interaction information and inputting human-vehicle interaction information; and establishing the output of the human-vehicle interaction LSTM neural network. According to the method, the advancing direction and speed of the pedestrian are selected as input, so that the influence of neighbors and vehicles on the current pedestrian motion is more intuitively shown. According to the method, influences of neighbor pedestrians and vehicles are introduced to serve as social information and serve as input together with pedestrian tracks, the neural network is constructed in a layered coding mode, the problem that the tracks of pedestrians are changed due to the influences of social factors can be solved, and the prediction precision is improved. According to the invention, a direction attention function is provided for distinguishing the influence of vehicles in different directions on pedestrians, so that the precision of social information is improved, and the pedestrian trajectory prediction precision is further improved.

Description

technical field [0001] The present invention relates to a pedestrian track prediction method, in particular to a LSTM (Long Short-term Memory Networks, LSTM for short) neural network prediction method based on human-vehicle interaction. Background technique [0002] Pedestrian trajectory prediction is of great significance in the research of autonomous driving technology, especially in high-density mixed traffic environment. In a mixed traffic environment, pedestrians, non-motorized vehicles and motorized vehicles and other "intelligent units" with observation, thinking, decision-making and action capabilities move in the same road space and interact and interact in the shared area. For a smart car driving in a mixed traffic environment, it is not enough to just avoid it when interacting, which will cause the vehicle to "hesitate" or "fall back and forth" for a long time, and the onlooker behavior brought about by its novelty is more likely Cause abnormal traffic flow. The...

AI Technical Summary

Problems solved by technology

However, the pure LSTM modeling method ignores the interaction between pedestrians and pedestrians, pedestrians and the surrounding environment, and cannot explain the trajectory changes caused by pedestrians in order to avoid collisions or form groups.

Images