Intelligent vehicle prediction control method based on visual spatial-temporal characteristics

Abstract

The invention discloses an intelligent vehicle prediction control method based on visual spatial-temporal characteristics. Firstly, a steering wheel angle prediction network is constructed, includinga spatial characteristic extraction network, N spatial-temporal characteristic extraction modules and a spatial-temporal characteristic map fusion prediction module, characteristic maps of different scales and different time steps are obtained by the spatial characteristic extraction network, the spatial-temporal characteristics are extracted from the characteristic map of each scale by the spatial-temporal characteristic extraction modules, then the spatial-temporal characteristic map fusion prediction module fuses the spatial-temporal characteristics of different scales to predict the steering wheel angle; and after the steering wheel angle prediction network is trained, a moment to be predicted is predicted, and exponential weighted average is performed on the predicted value of the steering wheel angle and the historical predicted value to obtain the final predicted value of the steering wheel angle. According to the method, the spatial-temporal information in the continuous imageframes can be effectively extracted, and the spatial-temporal information of different scales is fused together so that the prediction control precision of the intelligent vehicle is greatly improved.

Description

technical field [0001] The invention belongs to the technical field of intelligent vehicle control, and more particularly relates to a method for predicting and controlling an intelligent vehicle based on visual spatiotemporal features. Background technique [0002] The end-to-end decision-making method of smart vehicles means that the vehicle can automatically correct the deviation of the vehicle according to the situation it faces when driving in the lane. The traditional end-to-end decision-making method for smart cars generally requires the following steps: the sensor module composed of cameras obtains the image of the road ahead, sends the image to the perception module to detect the lane lines in the image, and then according to the lane lines, vehicle status, The relationship between the vehicle pose and the vehicle's driving direction calculates the number of steering wheel rotation angles required to maintain lane lines at the current moment. The end-to-end decisio...

AI Technical Summary

Problems solved by technology

Although this method improves the accuracy of the end-to-end decision-making network by using feature cascading, none of the methods mentioned in this method is reasonable, and in this method, the spatial feature Extraction and extraction of temporal features are still two separate processes

Images