This paper presents a robust method for close-range obstacle detection with arbitrarily aligned stereo cameras. System calibration is performed bymeans of a dense grid to remove perspective and lens distortion after a directmapping between image pixels and world points. Obstacle detection is based on the differences between left and right images after transformation phase and with a polar histogram, it is possible to detect vertical structures and to reject noise and small objects. Found objects’ world coordinates are transmitted via CAN bus; the driver can also be warned through an audio interface. The proposed algorithm can be useful in different automotive applications, requiring real-time segmentation without any assumption on background. Experimental results proved the system to be robust in several envitonmental conditions. In particular, the systemhas been tested to investigate presence of obstacles in blind spot areas around heavy goods vehicles (HGVs) and has beenmounted on three different prototypes at different heights.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below