Point Cloud Data Augmentation for Linear Assets

Machine learning algorithms are creating new approaches to address issues faced by the industry. These methods are data hungry for the right quality of data. Light Detection and Ranging (LiDAR) systems are becoming popular for collecting spatial data from vast areas. Processing of such large datasets and extraction of assets previously depended on tuning algorithms and development of features. Deep learning techniques free the user from the burden of generating and hand-designing features to extract assets of interest but require labelled data sets for training. Data augmentation is a popular technique for generating labelled data sets from available data and information. Data augmentation becomes the key to use machine learning by providing labelled point cloud data. Linear assets such as power lines, pipelines, railway tracks and roads extend over large areas and are designed as per set of specifications. Data augmentation for linear assets poses requirements set on specifications and simple data augmentation techniques do not fulfil the requirements. The objective of this paper is to explore available data augmentation techniques for image and point cloud and access their applicability to generate augmented point cloud data for linear assets specifically railways.