Indoor Point Cloud Registration Process Using ICP Algorithm

Abstract

In this paper, there are three major sections covered. The paper begins with the explanation of the Lidar sensor, which is a device that uses one or more lasers to measure distances from itself to objects in the world. Such a measurement can be represented as a point cloud, which is a set of (x,y,z) values for every surface that the laser(s) irradiate. Then, the theory of rigid transformations, including the covariance matrix, the translation matrix, and how the Singular Value Decomposition (SVD) can decompose the covariance matrix between two point-clouds to obtain the rotation matrix relating them. In addition, the concept of Iterative Closest Point (ICP) is explored and tested, along with the SVD algorithm, by using the Stanford bunny data sets. Finally, the ICP algorithm is used to combine two different Lidar scans of an office room together to see its effectiveness in a real-world environment.