One approach to detection and extraction of on-road obstacles based on image processing

Abstract

The process of automatic control and guidance of robots and autonomous vehicles is frequently based on the processing of image sequences obtained by a camera mounted on robot/vehicle. The first step in all algorithms of autonomous guidance of this type of devices consists in the analysis of scene contents and detection and extraction of obstacles that exist in the camera’s field of view. In the particular case of this research work, the extraction of “on-road” obstacles, which are generally moving, is done in order to track them from frame to frame and to adjust the device’s motion according to obstacles’ positions and relative velocity (incoming, outgoing, stationary ones). An initial segmentation oriented toward extraction of a road region is based on the Support Vector Machine (SVM) method of classification and learning. The following step consists in the extraction of “non-road” objects over the road area in order to mark the obstacles to be tracked later. The verification of this algorithm is done using typical scenes including the images of structured roads, urban environment and country roads.