Simultaneous Segmentation and Superquadrics Fitting in Laser-Range Data

Abstract

This paper presents a method for simultaneous segmentation and modeling of objects, detected in range data gathered by a laser scanner mounted onboard ground-robotic platforms. Superquadrics are used as model for both segmentation and object shape fitting. The proposed method, which we name Simultaneous Segmentation and Superquadrics Fitting, relies on a novel global objective function that accounts for the size of the object and the distance of range points, and for partial occlusions. Results on experimental 2-D range data, which are collected from indoor and outdoor environments, are qualitatively and quantitatively analyzed. Results are compared with those from popular and state-of-the-art segmentation methods. Moreover, we present results on 3-D data obtained from an in-house setup and also from a Velodyne LIDAR. This paper finds applications in areas of mobile robotics and autonomous vehicles, namely object detection, segmentation, and modeling.