REAL-TIME VISUAL PERCEPTION FOR TERRAIN MAPPING IN A WALKING ROBOT

This paper presents a new method for terrain map building with hardware-embedded stereovision. A real-time range data post-processing method is proposed to remove artefacts, and a method for building and updating the grid elevation map from point cloud data is presented. The relatively low computational load of the embedded stereovision and the presented post-processing method allows for real-time application. Experimental verification of the map building algorithm is provided. Also integration of the stereovision-based map with the more precise map constructed from laser data is described.