Online Detection and Tracking of Pipes During UAV Flight in Industrial Environments

Abstract

3D LiDaR-based perception has been used in robotics for obtaining accurate representations of the robot surroundings. As pipes are one of the most common objects in industrial environments, cylinder detection systems provide valuable information for robot navigation in industrial scenarios. However, most cylinder detection approaches using 3D LiDaRs suffer from high computational requirements and, therefore, their on-line implementation on real robots is limited for certain applications. This work proposes a computationally-light probabilistic approach to pipe detection and tracking suitable for aerial robot mapping and navigation. The proposed method was tested in both simulation and real scenarios. Through a combination of previous estimates and the localisation of the robot, the proposed approach is capable of reducing the computational cost of the RANSAC algorithm while keeping high detection accuracy.