In this paper we consider the problem of constructing a precise passability model by point cloud obtained from stereo cameras. To solve this problem, we used a hierarchical elevation map. The criterion for dividing cells into smaller ones was extended to take into account the cell's completeness with a limited field of view. The passability model was also supplemented with an algorithm for detecting step obstacles. The accuracy and completeness of the detection of obstacles and a free surface was calculated experimentally on dataset taken in real conditions. The result of the experiment shows that the proposed approach increases the number of detected obstacles without significant loss of accuracy.
Keywords: passability model, elevation map, obstacle detection, stereo vision, principal component analysis