An Advanced Driver-Assistance System: Using Mono-Camera Vision-Based Detection and Tracking for Control of an Autonomous Vehicle

Advanced driver assistance systems used to increase the safety of everyday transportation are increasingly common in today’s automotive industry. In conducting research for the Texas Department of Transportation’s Commercial Truck Platooning Project, an autonomous truck capable of communicating with a manually driven lead vehicle can follow at close distances to increase the fuel economy and safety of both vehicles. In the event of a communications failure, however, the following vehicle requires immediate human intervention to regain control of the vehicle. To increase the safety of these situations, a backup vision-based advanced driver assistance system comprised of a separate camera and computer was developed and successfully implemented in an autonomous vehicle. The system is comprised of a controller that takes inputs from a truck detection algorithm and a tracking algorithm that run in parallel at all times. The detection algorithm passes new detection information to the tracking algorithm when available. In each frame of incoming video, the tracking algorithm communicates the position of the lead vehicle to the controller. Based upon the size of the detection and the intrinsic properties of the camera, the longitudinal distance to the lead vehicle and the lateral error are calculated. The detection, tracking, and control algorithms were implemented in an autonomous car. A variety of lane changes and emergency stop maneuvers were successfully completed. GPS position information was gathered during testing to validate the accuracy of the algorithms. The results of the GPS information show that the tracking algorithm can encounter some error in estimating the separation at large following distances of more than 30 meters. Overall, the single camera system is capable of providing smooth and safe control in a variety of test scenarios