Autonomous Vehicle Tracking Control Using Deep Learning and Stereo Vision

Abstract

In this paper, a vehicle autonomous tracking control strategy is proposed through fusing neural-network based control, deep learning, stereo vision and Kalman filtering. In particular, a neural network controller is developed to utilize the vision and distance information and adjust the translational and rotational speeds of the follower vehicle so that it can track its leader autonomously. The SSD (Single Shot MultiBox Detector) deep learning technology is employed to detect the position of the leader vehicle visually, an image filtering algorithm based on the depth image is proposed, and a dual-Kalman filtering approach is presented to improve the reliability and speed of vision and distance measurements. The experimental results validate the effectiveness of the proposed strategy.