光电跟踪与测量技术研究室知识库

We propose to use shape descriptors and optimal search for fulfilling the pose estimation task based on optical measuring. Certain element of invariant moments is altered to meet the pose estimation requirements. Initial pose angles are acquired from the viewpoint-dependent appearance model of the object. 2-order normalized center moment is taken as the main feature descriptor and pattern search method is used as the search strategy. For simulation, a missile model of Milkshape 3D format is used. A model library of 1° resolution is created. Scale changing and noise adding are tested on 20 test images randomly selected from model library. For initial images, we can arrive at 1° precision. For scaled and noised images, even there appears some outlier, most of the items arrive around the true values. Results show that the method is practical. The method may get better results by improving the resolution of the model library.