Using principle component analysis to estimate geometric parameters from point cloud LIDAR data
Travis W. Sawyer,Andres Diaz,Esen Salcin,Jonathan S. Friedman
DOI: https://doi.org/10.1117/12.2585574
2021-04-12
Abstract:Light Detection and Ranging (LIDAR) is a popular sensing technique to measure static and dynamic objects with applications in many areas of defense technology including robotics, aircraft navigation and guidance systems, autonomous vehicles and aircraft landing systems, as well as tracking and measuring attitude of hypersonic objects. Despite widespread use of LIDAR to map out objects and environments, there remains a need for advanced analytic techniques to recover quantitative information about objects from LIDAR data, for example, the position and trajectory of a foreign object. One major class of LIDAR systems are those that produce so-called point-cloud data, which is a threedimensional sampling of a scene. Technical demands for extraction of geometric parameters from point-cloud spatial models are increasing as 3D LIDAR sensors and their application technology is continuously developed and popularized. While classical techniques for feature extraction and estimation exist, these existing techniques are currently inadequate to recover geometric parameters with desired accuracy for precision applications. To address this challenge, we developed an algorithm based on principal component analysis (PCA) to extract precise geometric parameters from LIDAR point-cloud data of objects including pitch, yaw, roll and xyz-position, as well as the rates of change of these parameters. We present the basis of this algorithm, as well as initial results using point cloud data of a rotating cylindrical object. The results suggest that PCA-based analysis could provide a robust and high precision approach for recovering object position and orientation, particularly when combined with other analytical approaches such as machine learning.