Real-time obstacle detection based on stereo vision for automotive applications

This paper presents a novel algorithm for on-road obstacle detection based on stereo cameras. The proposed algorithm significantly reduces the complexity disparity calculations involved when using a stereo vision technique. Many recent stereo-vision based obstacle detection systems require a dense disparity map and, then, locate the obstacles according to the depth information. However, calculating the correspondence for each pixel is very time consuming. In automotive applications object detection must be performed in real-time. The proposed algorithm uses given parameters of stereo cameras to determine the disparity search range of the pixels assuming all the pixels are on the road surface. According to the predefined search range for the road surface, the true disparities of obstacles are not included. Therefore, large errors will be introduced during block matching which indicate obstacle positions. This new system only consumes less than 10% of the traditional block-based disparity calculations. The core part of the proposed algorithm was implemented on the TI DM648 platform and achieved real-time performance.