摘要
研究了两幅深度图像的自动配准问题。在配准两幅深度图像时,基于多尺度空间理论得到两幅深度图像的特征集,然后利用该特征集进行深度图像匹配。具体过程为:首先为模型构造一个尺度空间,并寻找一个合适的尺度,计算深度图像各点的特征值;然后,利用该特征值对深度图像进行聚类分割;接着,在两幅深度图像得到的小分块中寻找对应分块;紧接着,使用RANSAC算法,根据三维几何信息的约束找出候选集中正确匹配分块和相对应的匹配点,并根据这些匹配点计算出两幅深度图像间的刚体置换矩阵;最后,使用改进的ICP算法优化这一结果。特别地,提出了一种行之有效的尺度空间构造方法,并提出了一种简单的顶点聚类算法,大大提高了聚类速度。
We have studied the automatic registration of two range-images. Based the theory of multi-scale space, we obtain the feature value of the two range-images, and then we match the two range-images using the feature values. Specific process: First, we construct a scale space for each range-image, and find a suitable scale to calculate the feature value of the points in the two range-images; then, take the segmentation using the feature values segmentation; then, search for the corresponding sub-block in the sub-block of the two range-images; then, take the RANSAC algorithm, in accordance with the binding of three-dimensional geometric information to find the right matching block in the candidate and the corresponding matching point, and using these two match points to calculate the rigid transform matrix of the two range-image; Finally, we use ICP algorithm to improve the results. Particularly, the paper presents an effective method of constructing the scale space, and a simple clustering algorithm, greatly improving the speed of clustering of the points.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2009年第S1期131-135,共5页
Journal of System Simulation
