结合纹理信息的深度像匹配

Range Images Registration Combined with Texture Information

利用图像的相关性在不同深度像对应的纹理图中自动选取特征点对;然后根据Hausdorff距离判断对应点对的有效性,并利用这些对应点对获得相应的深度点对,从而计算出2个深度像之间的初始位置变换关系;最后结合一种改进的带纹理的ICP算法实现深度像的精确配准.实验结果证明:该算法不仅适用于自由曲面的深度像匹配,而且适用于规则对称物体的匹配.

A method for the coarse registration of range images based on texture features is proposed. By the method, the feature pixel-pairs in two texture images are identified automatically by cross-correlation, and the validity-checking of these pixel-pairs is implemented through Hausdorff distance comparison. The correspondence between the texture image and range image helps to acquire the range point-pairs, and the initial transform of two range images is calculated by the least-square technique. With this initial transform, the fine registration is achieved by Texture ICP algorithm. Not only can this method deal with the registration of two free-form surfaces, but also the objects with highly symmetric shapes. The experiment demonstrates the efficiency of the algorithm.