A General-purpose Method to Extract Features from LIDAR Data

A novel,prior-knowledge independent and general-purpose feature detector for LIDAR(light detection and ranging) data is proposed to solve the feature detection problem in SLAM(simultaneous localization and mapping).The method adopts the Harris corner detector,a classic feature detection method from the image processing literature,specifically, the multi-scale Kanade-Tomasi corner detector,to extract features.The proposed method is capable of identifying stable and precise features from measurements at a variety of spatial scales,and produces feature descriptors and uncertainty estimates at the same time.The proposed method is applied to both simulated environments and standard datasets,including Victoria Park and Intel Research Center(both 2D),and the MIT DARPA Urban Challenge dataset(3D),and the experimental results strongly support that the proposed method can extract precise,highly repeatable and stable features from various environments.