Development of a reference software platform for the Velodyne VLP-16 LiDARS

Abstract: Recent advancements in sensor technologies have paved the way towards the realization of autonomous vehicles. The choice of sensors used in such vehicles plays a pivotal role in their successful operation under different conditions. The sensors used in the current generation semi-autonomous vehicles provide a reliable performance for normal use cases. However, their reliability significantly reduces for extreme use cases due to their sensing limitations such as, poor sensitivity to certain materials, limited range, restricted field of view (blind spots), and so on. Hence, a deeper understanding of the limitations of these sensors is required to characterize the sensor specifications required for future fully autonomous vehicles. This thesis work focuses on developing a reference software platform that captures the ground truth required for benchmarking the sensors used in current generation semi-autonomous vehicles. A modular, scalable software framework for the reference software platform called, LiDAR Software Platform is proposed, and four functions, namely point cloud fusion, point cloud mapping, point cloud obstacle detection, and point cloud map obstacle detection, based on the proposed framework are developed. The platform is developed using state-of-the-art Velodyne VLP-16 3D LiDARS (Light Detection and Ranging), the Robot Operating System (ROS), and the Point Cloud Library (PCL). A novel approach for improving the density of the point clouds by fusing the data from two Velodyne VLP-16 LiDARS is proposed, and successfully implemented, and a high density 3D point cloud map of the environment with centimeter accuracy is generated. Finally, the minimum point cloud density required for a robust detection of an obstacle at 50 meters from the truck is determined.