Research CUDA Simplifies Cloud Registration Algorithm Based On Ear Points

Anthropometry theory has pointed that the auricle shape which is unique and stable ofbiological information can be used for feature recognition. As in recent years, based on3Dlaser scanning data acquisition technology and digital geometry basic theory of mature, thethree-dimensional auricle scan point cloud data registration and its shape feature extractionmatching technology have become the hotspots in the field of international biometric.In the process of using three-dimensional laser scanner to obtain the auricle point clouddata, due to the limit of the angle, each scan can only get part of the auricle surface pointcloud data, it’s need to do multiple scans at multiple angles, and multiple scans data fusion inorder to get a complete registration. The critical time complexity of the algorithm ofregistration and registration accuracy decide auricle point cloud data accuracy. With theexpansion of the scale of point cloud data and improved registration accuracy requirements,the traditional serial registration algorithm obviously can not meet the real-time requirements.Firstly, based on the discrete curvature estimation and3D SIFT algorithm to extract afeature point of the3D auricle scan data, and then the non-feature points can be simplifiedusing the kd-tree, to thereby obtain the simplified geometric characteristics reserved auriclepoint cloud; then CUDA-based EM-ICP and Softassign algorithm parallel speedup. Thisalgorithm is fully considered in the simplified process, the geometric features, thus avoidingthe partial alignment of other defects, while the parallel EM-ICP Softassign algorithm, toensure that the efficiency of the alignment work.