中文摘要：

提出了一种基于局部描述符的物体识别算法．算法根据点云位置信息得到其矢量和曲率信息，根据形状索引提取特征点，在每个特征点根据矢量夹角把点云物体分割成不同的曲面片，每个矢量夹角曲面片通过一个二维直方图描述．该图显示了特征点与邻域之间法线矢量夹角对特征点法线矢量与特征点到邻域矢量之间夹角的出现频率．对于给定的一个物体，通过比对预测物体和模型物体的曲面片集描述，可得到潜在的对应模型物体，再通过迭代最近点算法，得到最终的识别结果．

英文摘要：

One novel local descriptor based on vector angle for 3D object recognition was studied. The information of KNN, vector and shape index are extracted from points＇ location information. The feature points are obtained based on shape index. With vector angle, surface points are segmented into patches centered in feature points. A vector angle surface patch descriptor is characterized by a 2D histogram. The 2D histogram shows the frequency of occurrence of the angles between the normal vector of reference feature point and that of its neighbors vs the angles between the normal vector of reference feature point and the vectors from the feature point to its neighbors. Finally, the patches descriptors of all the model objects are saved into database. Given a test object, a set of vector angle surface patches are created and compared with model patches in the database. Based on potential corresponding surface patches candidate models are hypothesized. After correspondence filtering of each pair of model and test object, verification is performed by running Iterative Closest Point （ICP）. The final identification result will be the pair of model and test object with the minimal RMS error. Experimental results with real images are presented to demonstrate and compare the effectiveness and efficiency to the proposed approach with the spin image and the tensor representation.