3D point cloud NDT odometry

Overview

Odometry from a lidar comes down to registering one scan against the next. The registration method decides how well it holds up.

For this Computer Vision course project I developed an odometry package in ROS using a Velodyne VLP-16. The algorithm uses the Normal Distributions Transform for scan matching between 3D point clouds.

The work followed on from a robotics internship at Tsuneishi Shipbuilding in Fukuyama, Japan, from May to June 2016, where I implemented ICP and NDT-based localization with the same VLP-16 sensor.

Tools

ROS, Velodyne VLP-16, Normal Distributions Transform, 3D point clouds.