3D feature points detection on sparse and non-uniform pointcloud for SLAM

In this paper, we propose a novel 3D feature point detection algorithm using Multiresolution Surface Variation (MSV). The proposed algorithm is used to extract 3D features from a cluttered, unstructured environment for use in realtime Simultaneous Localisation and Mapping (SLAM) algorithms running on a mobile robot. The salient feature of the proposed method is that, it can not only handle dense, uniform 3D point clouds (such as those obtained from Kinect or rotating 2D Lidar), but also (perhaps more importantly) handle sparse, non-uniform 3D point clouds (obtained from sensors such as 3D Lidar) and produce robust, repeatable key points that are specifically suitable for SLAM. The efficacy of the proposed method is evaluated using a dataset collected from a mobile robot with a 3D Velodyne Lidar (VLP-16) mounted on top.