Based On CUDA 's Ear Data Grid Merge Algorithm

Nowadays, biometric recognition technology has become one of the most popular personal identification technologies. Biometric recognition make use of the unique physical or behavior characteristic to identify the single individual, which is also of high stability, safety and reliability. Ear recognition, which uses the physiological position and structure characteristic, has become one of the most potential technologies in the field of biometric characteristic recognition, and it is of great significance in biometric recognition field and has drawn increasing attention of the public. In order to acquire complete data of ear, firstly we use 3D scanner to scan the ear from different angles. Then we use the collected data to make a registration. Finally, we merge the data after the registration step, and the scale of collected data will directly affect the efficiency of this step, which is the reason why we develop the GPU parallel technology to break through the limitation of data scale.This paper focus on how to effectively and accurately merge the multiple ear data model, and the main contents are as follows:Firstly, we use the 3D laser scanner to acquire three-dimensional human ear data, which includes the point cloud position information and the triangle mesh geometry information. Then, we use two-step iterative closest point algorithm to make ear point cloud data registration: Step 1, we adapt EM-ICP algorithm based on CUDA parallel mechanism to make the initial registration in order to make the ear point cloud data in the same direction to provide a favorable initial transform for following step. Step 2, we use the ICP algorithm for accurate registration of 3D ear data to get an accurate data for the later work. After registration step, we use the delete-patch merge method merge the mesh of registered ear data. By blocking, deleting the patch and stitching up the data from multiple directions of a single ear, we can finally acquire a complete, single 3D ear data. Meanwhile, we adapt CUDA to realize the parallel algorithm of the mesh-merging step in order to reduce the time complexity and speed up the algorithm. The experimental results show that, the merged 3D ear data performs an outstanding merging result and low time complexity.