A sliding window based feature extraction method with its application to 3D grid map fusion

Abstract

As the scale of the environment increases, the single robot simultaneous localization and mapping (SLAM) is not the best choice for building maps. In recent years, more and more researchers have focused on multi-robot SLAM. Multi-robot SLAM has the advantages of high efficiency, high precision of mapping and strong fault tolerance. However, multi-robot SLAM needs to solve the problem of map fusion. To solve this problem, this paper proposes an algorithm based on the maximum common subgraph for 3D grid map fusion, which does not require the pose relationship between robots. According to the characteristics of the indoor environment, the points where the three faces intersect are defined as corner points. Firstly, this paper proposes a sliding window-based feature point extraction method for 3D grid map. Moving the window to determine whether the point is a flat point, an edge point or a corner point according to the change in the number of occupied cubes in the window. Then we fit a plane to reject points that are not intersecting by three faces. Secondly, the backtracking method is used to find the maximum common subgraph. Finally, it calculates the transformation matrix according to the matched feature points to realize the fusion of 3D grid maps. The experimental results show that the proposed method can accurately fuse the 3D grid map with rotations.