Automated Labelling of 3D Point Cloud Data

Abstract

Semantic understanding of an environment is an important aspect of many computer vision and robotics applications. In order to acquire information for semantic labelling of an environment, many different sensors are being deployed in the research, but the use of Lidar and cameras have been extensive due to their portability. Information fusion of Lidar and camera provides a comprehensive way to the scene understanding, but for this labelled data is required. There is enough work done on labelling images but labelling point cloud data is still an open problem. In this work, a framework is devised for labelling point cloud through object identification and depth map estimation using the monocular camera in indoor and outdoor environment. The proposed framework does not require any training data for labelling point cloud data. Using this framework significant results with good accuracy are achieved on data set collected through Velodyne VLP-16 and camera.