The ability to detect and navigate drivable road surfaces is an important research area in autonomous navigation for use in autonomous vehicles. In this paper, a probabilistic computer vision algorithm for segmentation of tarred road surfaces is developed. Using a calibrated camera, a projection of a local obstacle map is then laid over the segmented image and an estimate is made of the likelihood of drivable region in each occupancy cell. The algorithm is both fast, can be implemented in real-time systems and robust, road surfaces are segmented well. The method was tested on a set of test images captured from a camera mounted on an autonomous vehicle. Good classification results are achieved, making it possible to use the algorithm and the resulting obstacle map in conjunction with global and local path planning algorithms to achieve autonomous navigation.
Reference:
Senekal, FP. 2009. Fast and robust road segmentation and obstacle map generation for autonomous navigation. 3rd Robotics and Mechatronics Symposium (ROBMECH 2009). 8-10 November 2009, CSIR International Convention Centre, Pretoria
Senekal, F. (2009). Fast and robust road segmentation and obstacle map generation for autonomous navigation. http://hdl.handle.net/10204/4149
Senekal, FP. "Fast and robust road segmentation and obstacle map generation for autonomous navigation." (2009): http://hdl.handle.net/10204/4149
Senekal F, Fast and robust road segmentation and obstacle map generation for autonomous navigation; 2009. http://hdl.handle.net/10204/4149 .