融合遗传算法和ICP的地面与车载激光点云配准

车载激光扫描可快速获取大场景点云，由于存在视场限制和遮挡，需地面激光点云作补充。车载与地面点云分别位于大地坐标和局部坐标系统，本文提出结合遗传算法（genetic algorithm，GA）和（iterative closed point，ICP）的自动点云配准方法以统一基准。ICP采用局部优化，效率较高，但依赖初始解；GA为全局优化方法，但效率低。融合策略为当GA配准趋于局部搜索时，采用ICP完成配准。GA配准以地面激光扫描仪内置GPS测量粗略位置限定优化搜索空间。为提高GA配准精度，提出了最大化归一化匹配分数之和（normalized sum of matching scores，NSMS）配准模型。实测数据试验验证了NSMS模型的有效性，GA配准均方根误差（root mean square error，RMSE）为1~5 cm；融合配准比GA配准效率高约50%。

Registration of TLS and MLS Point Cloud Combining Genetic Algorithm with ICP

Large scene point cloud can be quickly acquired by mobile laser scanning(MLS)technology, which needs to be supplemented by terrestrial laser scanning(TLS)point cloud because of limited field of view and occlusion.MLS and TLS point cloud are located in geodetic coordinate system and local coordinate system respectively.This paper proposes an automatic registration method combined genetic algorithm(GA)and iterative closed point ICP to achieve a uniform coordinate reference frame.The local optimizer is utilized in ICP.The efficiency of ICP is higher than that of GA registration,but it depends on a initial solution.GA is a global optimizer,but it's inefficient.The combining strategy is that ICP is enabled to complete the registration when the GA tends to local search.The rough position measured by a built-in GPS of a terrestrial laser scanner is used in the GA registration to limit its optimizing search space.To improve the GA registration accuracy,a maximum registration model called normalized sum of matching scores (NSMS)is presented.The results for measured data show that the NSMS model is effective,the root mean square error(RMSE)of GA registration is 1~5cm and the registration efficiency can be improved by about 50% combining GA with ICP.