Connected Components for a Fast and Robust 2D Lidar Data Segmentation

The paper presents a novel segmentation approach applied to a two-dimensional point-cloud extracted by a LIDAR device. The most common approaches perform well in outdoor environments where usually furniture and other objects are rather big and are composed of smooth surfaces. However, these methods fail to segment uneven, rough surfaces. In this paper we propose a novel range data segmentation algorithm that is based on the popular one-pass version of the Connected Components algorithm. Our algorithm outperforms most commonly used approaches, while keeping the low computational complexity. The algorithm is used as a part of control and perception system in our unmanned ground vehicle where real-time response time is required. We presented experimental results obtained indoors and outdoors. The latter experiment was conducted in the real test field while the vehicle was autonomously driven.