MACHINE LEARNING BASED INFORMATION RETRIEVAL FROM LiDAR AND UAV DATASETS

Abstract

Light detection and ranging (LiDAR) and unmanned aerial vehicle (UAV) are widespread now-a-days because these provide high-resolution datasets. They are known for delivering highly detailed information with the highest achievable accuracy. They are utilized in various applications, including hydrological, agricultural, smart city planning, geological, mining, building information modelling (BIM), e-commerce, product deliveries, etc. Despite numerous advantages and applications, their utilities are restricted due to various constraints, including complex data processing, security restrictions and the cost involved. In fact, current government policies are trying to ease their adoption at a larger scale; still, complex data processing is a great challenge. Machine learning (ML) falls under artificial intelligence (AI) and deals with developing algorithms that can learn the patterns of the data itself. The ability of ML to understand and find patterns in large datasets could be handy in processing large point cloud datasets obtained through LiDAR and UAV photogrammetry. This research work is focused on assessing the potential of ML algorithms for large-point cloud data classification. The present study has multiple objectives for which the descriptions are given below. This research focuses on identifying the prospective use of airborne LiDAR datasets in generating the terrain parameters. Aerial LiDAR point cloud datasets are known for their application in generating high-resolution digital elevation model (DEM), a basic requirement to extract/generate/derive other terrain parameters. This study estimates the digital surface model (DSM), bare earth model (BEM), aspect, slope, curvature and terrain ruggedness index (TRI) from the aerial LiDAR point cloud data. Further, these parameters are analyzed to measure the impact on the terrain characteristics and to highlight the potential of the airborne LiDAR data. The airborne LiDAR dataset used in this study is from the St. Petersburg region of Russia. For the accuracy assessment of the generated parameters, a ground-based survey is conducted to measure the ground control points using Differential Global Positioning System (DGPS) technique. They are used to assess the accuracy of the generated digital elevation model. An RMSE of 1.30 cm is obtained for the generated DSM with a maximum difference of 3.5 cm magnitude from the ground truth data. A more detailed analysis of the generated parameters is done to observe the effectiveness of aerial point cloud data for terrain parameter extraction.