|
Abstract:
|
In Computer Vision, 3D registration problem or aligning multiple 3D data sets to single coordinate system problem is fundamental and essential. Its applications vary from reconstructing 3D models to shapes matching, which can be applied to medical visualization, environment reconstruction or robot navigation. This problem has attracted many researchers’ attention. Many methodologies have been proposed: searching for global optimization transformation by finding correct correspondences, local optimization by minimizing an error metric that measures the closeness of two input data sets or combining both methods. In our thesis, we adapt the local optimization methodology, specifically, the Iterative Closest Point algorithm to solve the registration problem. However, as a local optimization process, the initial positions of point cloud data are crucial for algorithm performance. So, we proposed a semiautomatic process to aid this weakness and apply the algorithm to register a set of point cloud data. Three real world data sets have been used for testing this process and the result shows that the method is efficient and can be used for a real world application. |