Improved Loop Detection Method Based on ICP and NDT Registration Algorithm

Abstract

In the process of constructing a three-dimensional environment in a large scene, the problem of large accumulated errors is likely to occur, which makes it impossible to construct a globally consistent map. To solve this problem, this paper proposes a loop detection method based on NDT and ICP registration algorithms. The NDT algorithm is used for the initial registration, which can quickly reduce the number of the candidate loop frames and provide the initial pose. On the basis of the initial pose provided by the NDT algorithm, ICP algorithom is performed to determine the precise loopback frame. In this paper, the proposed loop detection algorithm is added to the SLAM framework, verified on the inspection robot platform, and compared with the LOAM and LeGO-LOAM algorithms. Experiments show that the proposed loop detection algorithm can effectively eliminate the cumulative error in the construction of the large environment, build a global consistency map, and have high realtime performance.