Geometric approach to registration and mapping with multi-layer LIDARs

Abstract

The advancement of autonomous navigation technology has enabled a wide range of applications such as autonomous driving, industrial automation, and remote sensing. One of the most important components of autonomous navigation technology is LI- DAR odometry and mapping. Odometry involves estimating the movement of a mo- bile platform, such as a self-driving car, in relation to its surroundings without using measurements from a fixed reference point. This is particularly important in situa- tions where a fixed reference point is not available, or when physical landmarks may be obscured by other objects. Even when a fixed reference point is present, odometry can still be used to improve position estimates by limiting the changes in position over a short period of time. odometry is essential for a wide range of mobile robotics applications, including self-driving cars operating in GPS-denied environments. In mobile robotics, both cameras and LIDAR sensors are often used for odometry. LIDAR sensors measure the time it takes for laser pulses to travel to and reflect off objects in the environment, allowing them to collect range and bearing information about the scene. Despite their higher cost, LIDAR sensors are sometimes preferred over cameras for certain applications because they are not affected by ambient light- ing conditions and do not suffer from glare or motion blur in low-light situations. Ad- ditionally, the direct range measurement capability of LIDAR sensors can simplify the odometry algorithm by eliminating the need to estimate distances from multiple sensor measurements. The main contribution in this thesis is features extraction for the LIDAR Odom- etry and Mapping (LOAM) algorithm, often rely on features extracted from point clouds. This thesis proposes a novel detection algorithm SKIP-3D (SKeleton Interest Point) for extraction of features namely edges and planner patches from multi-layer LIDAR scan. SKIP-3D make use of the organization of LIDAR measurements to search for salient points in each layer through an iterative bottum-up procedure. In the process it removes the low curvature points to find edges and classifies the points from point clouds acquired from different view points are associated and used for their alignment. Evaluation is carried out using the KITTI odometry metric on the KITTI odometry dataset as well as a dataset collected at University of Parma and demonstrated our method on several robotic datasets in both structured and unstruc- tured large, three dimensional environments. Experimental results showed that Fast LIDAR odometry and Mapping (F-LOAM) based on SKIP-3D feature extractor per- forms at least better than original F-LOAM feature extractor. Additionally, this thesis also addresses the issue of using heuristic procedures to exploit the organized point cloud structure of LIDARs, and presents alternative methods for improving feature point search. Presenting specific techniques for leveraging the structure of LIDAR point clouds, and introduces two key contributions in this area: efficient range search and nearest neighbor search. These methods allow for more effective exploitation of the organized structure of LIDAR point clouds. To summarize, the results of the evaluation indicate that incorporating geometric information can improve the performance of the method. However, there is poten- tial for further improvement through the development of a more sophisticated scene appearance model and a more effective application of this model.