Environment Recognition for Robot Navigation Using Image-based 3D Reconstruction with Terrain Parameter Identification

Abstract

Three dimensional data is useful to analyze traversability of unknown environment. Mobile robots equipped with camera and inertia sensor can be a solution for 3D construction of environment due to its lightness and cheapness in comparison with laser range finders. In this paper, a 3D-reconstruction-based terrain identification method is presented to predict vertical acceleration when running into an uneven terrains. Gaussian process (GP) was performed as approximation function to predict vertical acceleration using velocity of mobile robots, and terrains height and posture angle information. The results show Gaussian process can give good approximation to predict vertical acceleration with those terrains parameter, and investigated angle posture and height correlation for prediction.