近接点特徴量を用いた屋外環境におけるループ検出

抄録

This research proposes a loop-detection method that can be applied to time-series data retrieved from 3D LIDARs. Low-dimensional geometric features called proximity points are computed from 3D pointcloud data. A set of proximity points encapsulates the geometric layout of surrounding objects with respect to the observation point, and it provides a convenient means for measuring the similarity between two observations. The proposed loop-detection method, which is based on the Smith-Waterman algorithm, detects a loop as a pair of partial time series that exhibits high cumulative similarity score. The proposed method is tested on a data-set obtained in out-door environment.