A Study for Obstacle Determination fusing MMW Radar and Vision Sensor

This paper, on the basis of attention mechanism and driver's visual consciousness, presented a robust and adaptive approach for on-road obstacle boundary detection through fusing vision sensor and millimeter wave (MMW) radar in order to implement the navigation task, which is the sequent research work of radar and vision fusion for autonomous ground vehicle (AGV) navigation. It is the first most important issue for practically real-time precise guidance, navigation and control (GNC) of AGV. The novel exploration of prior knowledge about the features, shadow and edge or the position indicator lamps was discussed and employed with the attentive information from millimeter wave radar. An adaptive threshold algorithm based on the local significant peak and trough inside peaks determination (LSiPTiD) of the gray image histogram was provided in order to automatically judge the validity of obstacle shadow detection and detect the obstacle robustly. Vertical edge detection and possible intensity mutant were integrated with shadow and radar vision fusion point so as to recognize obstacle boundary correctly and robustly for further GNC task. In the end, through huge experiments of real on-road measured data, the algorithm was verified to be robust, adaptive and fast for real-time applications in natural road environment.