表 C011 共 2 頁 第 2 頁 十一、研究計畫中英文摘要：
(一) 計畫英文摘要。
Keyword: Vision-based measurement and registration system, robust-fast registration from different sources, evolutionary algorithm(EA), iterative closest point plus KD-tree method(ICP-KD-tree), space band-pass filter(SBF)
This paper will propose a vision-based measurement and registration technique. The fast and robust 3D geometric data measurement and registration strategy dedicated for image-guided applications. The 3D geometric data from different sources can not be matched completely and these data points include a lot of distinct parts.
The registration scheme is composed of a coarse transformation stage and a fine-tuning stage. In the first stage, the proposed space band-pass filter (SBF) is used to reduce the data amount of template 3D image, the higher frequency’s part of the huge reference image data is searched and reserved by SBF, and evolutionary computation is implemented to find optimal initial pose for the ICP (iterative closest point) plus KD-tree (k-dimensional tree) scheme. In the second stage, its fine matching process similar to the original ICP approach plus K-D tree towards the end of the minimization of mean square error (MSE).
In the other hand, a high-accuracy 3D measurement system will be purposed for free points by using multi-camera stereo vision; furthermore, the system will reconstruct 3D surfaces of free form objects by smooth interpolation model. An accurate 3D object will provide essential information to be used in a variety of computer vision applications.
The multi-camera-based approach with structured light projection reconstructs unconstrained corresponding points on 3D objects to facilitate the registration. Using this projector, the pattern’s designs of structure projection can be made by user to be able to generate any key-point or SIFT (scale invariant feature transform) feature of target. And, all key-points on the 3D surfaces of free form objects with the patterns will be projected on observed image planes, and they will be detected by POC (phase-only correlation) method and SIFT approach. Then, according to the geometry relationship between these cameras, all of the key-points will be used to reconstruct the 3D objects by the smooth and modeling method. Finally, the proposed method will be tested and implemented through the registration application of the actual 3D data.
