Efficient Planar Surface-Based 3D Mapping Method for Mobile Robots Using Stereo Vision

Environmental mapping plays an important role in the field of robotics. Conventional voxel-based occupancy grid models largely reduce system mapping efficiency for huge quantities of grid cells. This paper presents an efficient method of 3D grid modeling using stereo vision based on planar surfaces. This method first uses matching key feature points with a multi-random sample consensus algorithm to estimate plane parameters and then clusters pre-processed point cloud data located on the same plane. Next, a split and combining algorithm is used to generate 3D planar grid approximation representations of the environment. The occupancy probabilities of grid cells are estimated and updated by using the Kullback-Leibler divergence. Finally, a series of experiments including map qualitative analysis and performance tests, are adopted to evaluate the presented method in indoor and outdoor environments. The results of the experiments and performance evaluation illustrate the capabilities of our approach in generating efficient 3D maps.