Robust Obstacle Detection in General Road Environment Based on Road Extraction and Pose Estimation

Understanding the general road environment is a vital task for obstacle detection in complicated situations. That task is easier to perform for highway environments than for general roads because road environments are well-established in highways and obstacle classes are limited. On the other hand, general roads are not always well-established and various small obstacles, as well as larger ones, must be detected. For the purpose of discerning obstacles and road patterns, it is important to determine the relative positions of the camera and the road surface. This paper presents an efficient solution using a stereo-vision-based obstacle detection method for general roads. The relative position is estimated dynamically even without any clear lane markings. Additionally, obstacles are detected without applying explicit models. We present experimental results to demonstrate the effectiveness of our proposed method under various conditions