Abstract
This paper presents a method of correcting distortion in 3D laser-scan data from in-vehicle multilayer laser scanner. A robot identifies its own 3D pose (position and attitude) in a laser-scan period using the normal-distributions transform (NDT) scan-matching method. Based on the pose information, the robot's pose in a period shorter than scan period under the assumption that the robot moves at almost constant linear and turning velocities. The estimated pose of the robot is applied to map laserscan data onto a world coordinate frame. Subsequently, the robot again identifies its own 3D pose from the mapped scan data using NDT scan matching. This iterative process enables the robot to correct the distortion of laser-scan date and accurately map the laser-scan data onto the world coordinate frame. Two methods for correcting the laser-scan data are presented: linear-interpolation based and Kalman-filter based methods. The former applies the extrapolation and interpolation to estimate the vehicle pose, while the latter applies the prediction, estimation, and smoothing. Experimental results of mapping a signal light in a road environment show the performance of the proposed two methods.
