Point cloud adaptive simplification of feature extraction

Point cloud data, as a kind of three-dimensional information reflecting the object shape, have quite a large amount of original data, so if directly operating on excessive data, it will affect subsequent work such as point clouds reconstruction, etc. This paper proposes a novel adaptive simplification algorithm for point cloud feature extraction. First, space should be divided with respect to the original point cloud, and then k neighborhood of the point should be built, and feature parameters should be set up, and then feature analysis should be conducted, and finally information and data of different parts should be identified. Then, for the planar data, the boundary is detected and extracted and the remaining parts are simplified. Finally, for the nonplanar data, the feature is extracted and then simplifications are implemented in varying degrees according to different curvatures. Experiments show that it takes no more than several seconds to process a point cloud model with almost a million points. Simplification proportion can reach above 90%, and the error corresponding to original data is smaller: the average deviation of the planar data is less than 0.02 mm before and after simplification, with a small fluctuation at 0.0057 mm; the average deviation of the nonplanar data is likely to fluctuate around 0.08 mm and the difference is only 0.0003 mm before and after simplification, guaranteeing the simplification accuracy. Therefore, the data processed by proposed algorithm can display the object shape better. © 2017, Science Press. All right reserved.