Vision-Based Detection, Decision and Control for Autonomous Vehicles in AD Chauffeur

For autonomous vehicles, camera is a kind of commonly applied sensor. A vision-based scheme involving object detection, motion decision and tracking control is proposed in this paper for autonomous vehicles running on highways. A YOLO-DeepSORT network and a LaneNet network are adopted to realize object and lane detection. The decision task is carried out based on a designed finite state machine. To stabilize the tracking error, an enhanced pure-pursuit controller associated with lanes and an integral-separated PI controller are developed for the autonomous vehicle. At last, AD Chauffeur, a cloud simulation platform is utilized to verify the effectiveness, which is developed by Automotive Data of China (Tianjin) Co., Ltd. The experiment results not only verify that the performance of our vision-based strategy but also illustrate the functions of AD Chauffeur platform.