Fast Occupancy Grid Filtering Using Grid Cell Clusters From LIDAR and Stereo Vision Sensor Data

Occupancy grid filtering is important for navigation and localization in dynamic environments. The techniques that have been proposed in the past research, such as the Bayesian occupancy filter and the sequential Monte Carlo Bayesian occupancy filter, involve the independent consideration of all grid cells and, thus, are inaccurate and computationally expensive. To solve this problem, we propose a fast occupancy grid filtering method that uses a grid state map. We apply a superpixel-based clustering concept called a grid cell cluster (GCC) to 3D particles measured using light detection and ranging and dense depth maps extracted from a stereo vision sensor. The GCC locates a connection among similar data items according to certain criteria. In our proposed method, a GCC consisting of several grid cells has one dynamic state. Thus, we can reduce the computational cost and the number of computational errors by using the same prediction matrices is generated from the grid cells belonging to an object. Our method was evaluated on the KITTI benchmark data set. The results showed that the proposed method was faster (by approximately 38.9%) than Mekahanachas method and approximately 12% more accurate (on the intersection-over-union metric) than prevalent techniques.