Mapping and 3D reconstruction based on lidar

Abstract

A 3D reconstruction and mapping framework aided by nonlinear filter optimization is introduced in this thesis project. A pipeline consisting of point cloud preprossesing, followed by a robust and rapid non-feature-based normal distributions transform (NDT) registration algorithm was developed for accurate mapping. An unscented kalman filter (UKF) solution to fuse inertial measurement unit (IMU) and global navigation satellite system (GNSS) measurements with augmented continuous turn rate velocity (CTRV) magnitude model is considered to obtain continuous six degree of freedom (6-DOF) pose estimation for accurate localization of the agent. The primary motivation for developing a high-quality mapping and localization system is because they play a key role in advancing towards an autonomous vehicle. Due to the lack of synchronized public datasets with IMU, GNSS, and light detection and ranging (lidar) measurements, the initial implementation of the mapping solution was tested over outdoor dataset from cars. The final phase of the project development is aimed at testing and tuning the program for the custom maritime dataset.