Simultaneous registration of multi-view range images with adaptive kernel density estimation

3D surface registration can be considered one of the crucial stages of reconstructing 3D objects from depth sensor data. Aligning pairs of surfaces is a well studied problem that has resulted in fast and usually reliable algorithms addressing the task. The generalised problem of globally aligning multiple surfaces is a more complex task that has received less attention yet remains a fundamental part of extracting a model from multiple 3D surface measurements for most useful applications. In this paper, we propose a novel approach for the global registration of depth sensor data, represented by multiple dense point clouds. Point correspondences between scans and view order are unknown. Given many partial views, we estimate a kernel-based density function of the point data to determine an accurate approximation of the
sampled surface. We dene an energy function which implicitly considers all viewpoints simultaneously. We use this density to guide an energy minimisation in the transform space, aligning all partial views robustly. We evaluate this strategy quantitatively on synthetic and range sensor data where we nd that we have competitive registration accuracy through comprehensive experiments that compare our approach with existing frameworks for this task.