Robust registration of 3D point clouds using GA with adaptable boundary constraints

Document Type : Original Article

Authors

Beijing University of Aeronautics and Astronautics, Beijing, China.

Abstract

Registration of 3D point clouds is an important task for many different applications today, like reverse engineering, medical imaging, remote sensing, robotics and automation in general. In this paper we explore a new algorithm for rough registration by using the genetic algorithm (GA) which has showed acceptable results in a reasonable amount of computational time. The algorithm is based on the Median Squared Error (MSE) as the fitness function for the GA, and a threshold that defines the maximum distance between corresponding pair of points to be considered inliers. One of the contributions in this paper is a new scheme for adaptable boundary constrains in the algorithm, which makes the global minimum detection faster than the traditional GAs implementations. Another contribution is the separation of the chromosome parameter, i.e. for the first steps of the algorithm we have separated the chromosome encoding into two groups; three genes representing the translation vector (Tx, Ty and Tz) and three genes representing the rotation matrices (Rx, Ry and Rz). Finally, experimental results are presented and discussed using this algorithm.

Keywords

Files

Share

How to cite

Statistics