Lin Hongbin

Feature preserving holes filling of scattered point cloud based on tensor voting

This paper presented a novel holes filling method for the scattered point cloud, particularly for recovering the missing points at featured curves or corners. Firstly, hole-boundary point and feature point were extracted to determine the incomplete feature curves. According to these two point sets, the holes were categorized into two types, featured holes and non-featured holes. Secondly, tensor voting was used to restore the missing feature points of the incomplete feature curves. Accordingly, original holes were divided into sub-holes without missing feature. Finally, all the non-featured holes were filled by the tensor voting. Experimental results show that the proposed method could accurately and effectively restore the original shape of the hole including the sharp features.

Research on a novel 3D point cloud robust registration algorithm

This paper present a novel algorithm for robust registration of 3D point cloud. It significantly increases the robustness against outliers by using the multivariate Epanechnikov kernel, which changes the registration procedure from exponential minimization procedure to polynomial minimization procedure. It reduces the computation complexity by using a new cost function. Because the analytical expression of the cost function is available, the registration procedure can converge in very little iteration. It also improves the flexibility of the algorithm by using the new parameterized cost function, which allows us to adjust the performance of the algorithm between robustness and efficiency.