Data reduction for long-term mapping using occupancy grid map with observation frequency

This paper presents a method of data reduction for long-term robotic mapping using a 2D laser scanner. The proposed method introduces observation frequency into the occupancy grid map, and calculates the importance of each scan through the ray-casting algorithm. The importance of a scan is high if the scan observes many occupancy cells which are new in terms of the observation frequency. Also, the method calculates the coverage of a subset of scans to examine how the subset covers the map well. Then, the method determines the threshold of the scan importance according to the designated coverage, and removes the scans which are less important than the threshold. The reduced laser scans improve computation time and memory consumption for pose adjustment and map reconstruction in loop closure. Experiments using real-world data show the proposed method effectively reduces data size and computation time in robotic mapping.