关键词: 物体识别, 点云, 测地线, 指纹, 局部描述符

Abstract: A 3D points cloud object recognition algorithm was proposed based on local descriptor. The vector and shape index value of the points cloud were calculated, and then according to the shape index, feature points were extracted. Through the geodesic distance and vector angle, points cloud was segmented into different patches centered on feature points. Through the geodesic partition on each patch, 3D geodesic-style concentric circles were got. Thus the description of 3D objects could be transformed into two 2D curves: the normal vector curves and the Euclidean distance curves, then one model objects database would be established. Through comparison of the descriptions with the model database, some potential recognition results of a given object could be found. With the final iterative closest point algorithm, the final recognition result could be determined. The experimental results in real objects demonstrate the effectiveness of the proposed algorithm.

Key words: object recognition, points cloud, geodesic curve, fingerprint, local descriptor