Abstract:  This paper proposes an improved ICP (iterative closest point) algorithm based on adaptive optimal thres-hold to align two shape models. These shape models can either be partial or complete. First, the models are pre-aligned using PCA (principal component analysis) roughly, then their size are adjusted using a three-dimensional scaling transformation. Second, the nearest neighbor search is carried out using KD-tree. This improves the speed of searching the corresponding points. Using ICP, the aligning error of the source and target model is calculated with respect to a set of threshold, then the threshold K_best corresponding to the minimum error is recorded. And then, K_best is taken as a new threshold to align the two models using ICP again. Finally, an algorithm for detecting coincidence degree      between 3D shapes is executed, thereafter, the source model is adjusted according to the resulting coincidence degree. The experimental results show that the proposed algorithm can align two models, especially for partial and complete models and clearly improves the accuracy of ICP algorithm.

Key words: model registration, principal component analysis, iterative closest point

摘要： 针对整体与部分3D模型间的配准问题，提出了一种基于自适应最优阈值的迭代最近点（iterative closest point，ICP）算法。首先使用主成分分析法将模型进行初始配准，并使用三维缩放变换调整模型的大小；然后采用KD-tree进行最近邻搜索以提高对应点的查找速度，计算在不同的阈值下对两模型执行ICP算法的配准误差，并记录误差最小时所对应的阈值K_best；再以K_best为阈值重新对两模型执行ICP算法，将目标模型和源模型配准；最后执行三维目标重合度检测算法，根据重合度再进行最后的反转调整。实验结果表明，改进的ICP算法既能配准整体与部分模型，也适用于两个完整模型间的配准，提高了ICP算法的精确度。

关键词: 模型配准, 主成分分析法, 迭代最近点算法