Research On 3D Model Reconstruction Based On Time-of-Flight Camera

With the development of computer technology, 3D vision technology has been used more and more widely in scientific computing visualization, education, military, and entertainment et al. Acquiring 3D models is one of the basic tasks in 3D vision technology. Time of Flight method is a common way for acquiring depth information of object surfaces. But the captured data of Time-of-Flight (ToF) camera can't be used directly in 3D reconstruction due to the low image resolution and high noise level. Therefore, this paper introduces some research to make these data can be used in 3D model reconstruction.1. In this paper, we do some researches on 3D model reconstruction based on range scans. Such as point cloud triangularization, bilateral filtering, median filtering, object segmentation, overlap detection and ICP algorithm etc. After summarize these algorithms, a method for low quality data processing is proposed. After this process, the data can meet basically the requirements of 3D reconstruction.2. A method for the pre-processed data is proposed on the basis of the high quality data reconstruction algorithm and characteristics of ToF camera. This method employs the iterative closest point algorithm and the point cloud noise processing to align range scans. Then it employs the Poisson surface reconstruction algorithm to reconstruct rigid objects models. The method not only stands the noise in the data, but also avoids the error accumulation of continuous multi-frame alignment effectively.3. Based on the rigid alignment algorithm, in order to put up with the motion blur and non-rigid deformation during the global non-rigid shape alignment, an algorithm for non-rigid method of reconstruction is proposed. Firstly, the captured data from different views are segmented and denoised. Secondly, the meshes are aligned roughly using rigid alignment, and feature frames are selected to reduce motion blur and redundancy. Then the feature frames are further aligned using global non-rigid alignment to deal with misalignment of static object with non-rigid errors and human body under low-frequency non-rigid deformation. Finally, a qualified mesh is generated by using Poisson surface reconstruction. The experiments show that the algorithm can improve the quality of captured depth data and extend the typical application of ToF camera.