Research On Point Cloud Registration Algorithm Based On Improved GA-SA

This paper mainly studies a point cloud registration algorithm improvement based on the fusion of multiple optimization algorithms.The optimal solution is directly searched in the solution space of the registration conversion parameters without the process of feature extraction and feature matching.In the process of point cloud coarse registration,the search efficiency,convergence accuracy and registration success rate of the optimization algorithm have been studied,and the related algorithms have been improved.Aiming at the problem of insufficient precision of the traditional ICP fine registration,the ICP is improved through the point-to-surface closest point query and setting the error-corresponding relationship elimination ratio method,which effectively improves the precision of the fine registration.Finally,the experimental verification was carried out through Stanford University point cloud data and measured point cloud data.The work on the paper is as follows:(1)In view of the defects of the root mean square error function(RMSE),which is commonly used for point cloud registration accuracy evaluation,large amount of calculation and poor global evaluation performance,a simplified and normalized root mean square error function(s NRMSE)is proposed.The point cloud data and calculation amount are greatly reduced,and the normal vector is added to assist in evaluating the distance value,so that the actual accuracy between the point clouds can be correctly evaluated for symmetry.(2)In view of the problem that genetic algorithm(GA)is prone to fall into premature convergence during global search,a GA algorithm incorporating immune operators and catastrophe operators,namely,immune genetic algorithm with catastrophes(IGAC)is proposed.IGAC can guarantee the diversity of the population in the optimization process,and force the update of the population distribution when the population appears to converge prematurely.Experiments prove that IGAC effectively suppresses the stagnation phenomenon,reduces the number of invalid iterations,and improves the global convergence performance.(3)In view of the problem that the simulated annealing algorithm(SA)is inefficient and insufficient in searching in the multi-dimensional solution space,according to the mathematical relationship between the conversion coordinates and the conversion parameters,a SA cycle optimization algorithm that introduces the long beetle search algorithm(BAS)into SA is proposed.The experiment proves that the SA loop optimization algorithm can efficiently and fully complete the local search of the conversion parameters,and has strong stability.(4)Considering the good characteristics of IGAC in global search and SA cycle optimization in local search,the IGAC-SA fusion strategy is proposed to achieve coarse point cloud registration: first,through the center of gravity and SA translation cycle optimization,the initial relative position of the point cloud is simply adjusted to reduce GA search interval;global search is performed based on IGAC and converges to the global extreme point;finally,the SA loop is optimized,and the translation and rotation parameters are optimized at the same time to converge to the local extreme point,that is,the optimal registration conversion parameter is obtained.Experiments show that the IGAC-SA hybrid strategy has higher convergence accuracy,faster search speed and higher registration success rate.