Evaluation of point cloud registration using Monte Carlo method

Supervision and control are one of the most important steps while executing a construction project and their automation remains an area of growing interest. LiDAR systems provide accurate point clouds with geometric information that can help to improve the automation of survey control. Alignment of the point clouds acquired from a number of scan positions is a fundamental issue regarding surveying accuracy and is frequently carried out in two steps: coarse and fine registration. Fine registration can be achieved automatically by means of an Iterative Closest Point (ICP) procedure. This work presents a Monte Carlo based method to quantify the reliability of a coarse registration step that would enable the automation of the alignment. The method consists of verifying the tolerance of a particular ICP implementation to coarse registration errors. Results show that the ICP alignment used works fine when coarse registration errors are lower than 18° for rotations and 1 m for translations. These values were similar for four case studies analysed. Quantifying these limits is crucial for operations such as robotic stop & go surveying, where coarse alignment is based on Simultaneous Location and Mapping (SLAM) and fine alignment is achieved through ICP.