基于深度传感器的多视角点云配准研究

Research on registration of multi-view point cloud based on depth sensor

为了解决大尺寸对称物体在多视角配准过程中出现的误匹配点对和累计误差问题,提出了一种基于深度传感器的多视角点云配准算法。首先,使用深度传感器获取目标物体不同视角下的多片点云并进行预处理,对物体单侧相邻点云采用超四点快速鲁棒匹配算法(Super 4-points congruent sets,Super4PCS)进行粗配准,利用改进的点到平面ICP算法去除误匹配点对并进行精配准,之后将左右两部分的点云拼接,从而获取完整的三维点云模型。最后,针对多视角配准出现的累计误差问题,提出了一种全局优化方法从而减少累计误差。实验结果证明所提方法可以精准地完成多视角点云配准,获得准确的三维点云模型。

To address the issues of mismatched point pairs and cumulative errors encountered during the multiview point registration of large-scale symmetrical objects,a multi-view point cloud registration algorithm based on depth sensor is proposed.Firstly,the proposed approach leverages depth sensors to capture multiple point clouds of the target object from various viewpoints,which are then subjected to a series of preprocessing steps.To achieve coarse registration,the Super 4-points congruent sets(Super4PCS)is employed specifically for adjacent point clouds on one side of the object.Subsequently,an enhanced point-to-plane ICP algorithm is utilized to refine the registration by eliminating erroneous point pairs.The resulting refined point clouds from the left and right sides are seamlessly combined,thereby generating a comprehensive 3D point cloud model.Furthermore,to mitigate the issue of cumulative errors arising from the multi-view registration process,a global optimization technique is introduced.Experimental evaluations demonstrate the effectiveness and accuracy of the proposed method in achieving precise multi-view point cloud registration and generating highly accurate 3D point cloud models.