Obstacle avoidance using stereo-based generic obstacle tracking

摘要

Obstacle detection and tracking is a fundamental task for several Advanced Driver Assistance Systems (ADAS) and self-driving cars. Several approaches have been presented in the literature in the last years and many of them are based on visual sensors. In this paper we propose an approach that uses only stereo cameras to detect and track obstacles and compute visual odometry to get the vehicle's ego-velocity. This information is used to compute a safe velocity to travel avoiding collisions with any detected objects. This is an ongoing work and its main contribution is not requiring a full dense disparity map to detect and track obstacles, improving efficiency. It also does not require a previous training step.