In this paper, we propose a novel framework for detecting and filling missing regions in point cloud data. We propose to investigate the properties of point cloud data and develop a framework that can detect boundaries of intricate holes in the point cloud data of complex shapes and fill the hole consequently. 'The holes in point cloud data are caused owing to many reasons like reflectance, transparency, occlusions etc. Detecting holes in point cloud data is a non-trivial task, since point cloud data is unstructured and comes with no adjacency/connectivity information. We propose a Centroid-Shift algorithm that exploits the distance of cluster centroid from the member points to detect the boundaries of holes, further we propose a polynomial surface fit framework to accurately fill the missing regions/holes without losing the original shape attributes of the objects. We demonstrate our framework on popular 3D objects, we provide qualitative results and report RMS error as the evaluation metric to measure the effectiveness of the hole-filling algorithm.
展开▼
