果实点云配准中的最优变换

Optimal transformation in fruit scans registration

为研究不同几何参数对基于特征相邻果实点云之间配准效率和准确度的影响,并进一步优化基于点元扫描片之间成对配准,利用两个坐标系之间的变换,提出了一种新的、直接由位置和法向量决定的最佳几何描述元,并基于该描述元通过消除冗余变换降低计算复杂性、简化变换过程,实现最优变换。对扫描得到的5组苹果片进行配准测试,结果表明:法向点元具有刚体变换不变性和噪声不敏感性,区分度高、复杂度低,将变换中所需的同名点由5对减少到1~3对,降低了搜索多组相应的描述元对的复杂性;最优变换求解过程复杂度低(仅需3次求模、4个三角函数,7次矩阵乘法)、准确度高(误差与法向点元对数量无关),缓解了获取变换参数依赖多个线性方程组的复杂求解现象。

For the study of influence of different geometric parameters on the efficiency and accuracy of feature based registration between fruit scans,and further optimizing the surfel based pairwise registration,this paper proposes a new, position-normal directly dominated optimum geometry descriptor,and make the transformation optimization by eliminating redundant transform to reduce the computational complexity and simplifying the transformation based on the proposed operator. By applying registration algorithm on 5 groups of apple scans,the results show that: the position-normal based low-dimensional geometry descriptor,normal-surfel,reduces the correspondences from 3 ~ 5pairs to 1 pair and avoids complex iterative process of searching for descriptor pairs,due to the invariance to rigid body transformation, noise insensitivity, high discrimination, and lowcomplexity; the optimal transformation and solution process only needs thrice module,4 times triangle function,7 times matrix multiplication,which results in high accuracy and avoids the complex calculation process of solving numerous linear equations for transform parameters.