Study Of Registration For 3D Point Clouds Based On Evolutionary Algorithms

Laser 3D scanning technology has been a kind of fast and efficient means of 3D digital method.It has been widely applied in surface topography measurement,garment manufacturing,reverse engineering,film and television entertainment,human body engineering design and so on.The existing methods of 3D point clouds data acquisition and data processing efficiency are expected to reach a higher level.The development of efficient point clouds registration algorithms has become the hotspot of academic research.Surrounding on the objective,the major work and innovation are as follows:1.We have done research on extracting feature points in large points quickly and accurately.This paper analyzed and compared curvature feature points extraction,random sampling,KPQ-points extraction,ISS feature points extraction.We propose an improved ISS algorithm adopting a neighbour points radius constraint strategy,make it suitable for general point cloud models.2.We analyzed the fitness function of the median distance between corresponding points which is used to describe the accuracy of image registration and proved the swarm intelligence algorithm can be used to optimize the function.We have realized the function of Particle Swarm Optimization,the Biogeography-based optimization,the Artificial Bee Colony Algorithm.3.Based on the researches above,the complete registration algorithm of point clouds has been developed by Matlab.Through the registration results of ideal models and real human model,the performance of different algorithms has been compared.We also obtained the relationship between over-lapping rate and registration accuracy.