【摘要】 针对传统点云配准中存在配准精度低、耗时长的问题，提出一种邻域多维度特征点结合相关熵模型的点云配准方法。首先根据邻域点的加权投影信息、表面曲率和法向量夹角提取特征点；其次用二值化的方向直方图描述子(B-SHOT)进行特征描述与匹配，然后利用刚性距离约束剔除误匹配，并通过随机采样一致性算法获取初始变换矩阵；在精配准阶段，以点到面的距离为准则双向搜索对应点，并通过多种几何特征约束剔除误匹配点对，最后迭代最大相关熵模型的目标误差函数完成精配准。实验结果表明，本文算法比迭代最近点算法(ICP)的配准精度提高了15%～97%、配准效率提高了约90%。更多还原

【Abstract】 Aiming at the problems of low registration accuracy and time consuming in traditional point cloud registration, a point cloud registration method based on neighborhood multi-dimensional feature points and correntropy model is proposed.Firstly, the feature points are extracted according to the weighted projection information, surface curvature and normal vector angle of the neighboring points.Secondly, the binary signature of histogram of orientations(B-SHOT)is used for feature description and matching, then the rigid distance constraint is used to eliminate the mismatches, and the initial transformation matrix is obtained through the random sampling consistency algorithm.In the fine registration stage, the corresponding points are searched in both directions based on the point-to-face distance, and the mismatches are eliminated through a variety of geometric feature constraints.Finally, the target error function of the maximum correlation entropy model is iterated to complete the precision registration.The experimental results show that the registration accuracy of this algorithm is improved by 15 %～97 % and the registration efficiency is improved by 90 % over the iterative closest point algorithm(ICP).更多还原