基于混合遗传算法和点面距离测度的深度像配准

Range image registration using hybrid genetic algorithm and point-to-plane distance-based measure

提出一种利用改进的遗传算法和点面距离作为误差测度的深度像精确配准算法。与现有ICP框架下的迭代算法不同,将深度像配准视为高维空间的一个优化问题,通过在遗传算法中加入退火选择、爬山法以及参数空间的动态退化来加速寻找最优的位置转换关系。同时,采用一种新的基于点面距离的适应函数来计算配准误差,使得算法具有更强的鲁棒性。实验结果表明,该算法不需要初始的运动参数估计,具有较高的配准精度,收敛速度快且抗噪声能力强。

This paper presented a novel approach for precise registration of range images pair with an improved genetic algorithm（GA） and a new error metric based on the point-to-plane distance. Different to the existed ICP methods, this approaeh formulated the surface registration as a high dimensional optimization problem. Then combined the strategy of simulated annealing（SA） selection, hill-climbing and dynamic parametric space degeneration into a GA to offer much faster convergence and more precise registration. At the same time, employed a new measure based on the point-to-plane distance as fitness function to evaluate the alignment error, which made the approach more robust. A number of experiments demonstrate that the presented method is insensitive to noises as well as the initial pose estimation and has high precision and fast convergence.