Linewise Non-Rigid Point Cloud Registration

Robots are usually equipped with 3D range sensors such as laser line scanners (LLSs) or lidars. These acquire a full scan in by manner while the robot is motion. All lines can be referred to common coordinate frame using data from inertial sensors. However, errors noisy measurements and inaccuracies extrinsic parameters between scanner also projected onto shared frame. This causes deformation final containing all lines, which known motion distortion. Rigid point cloud registration methods like ICP therefore not well suited for distorted scans. In this paper we present non-rigid method that finds rigid transformation applied each order match an existing model. We fully leverage continuous relatively smooth respect scanning time formulate our reducing computational complexity improving accuracy. use synthetic real benchmark against state-of-the-art method. Finally, source code algorithm made publicly available.