Segmentation of Complex Object Based on A Standardization Coding Method

Abstract

Accurate segmentation of complex objects from point cloud data is a useful but difficult task in many applications. In order to achieve this goal, this paper first proposes a standardization coding method of 3D point cloud data by establishing a set of coordinate systems as the benchmark for feature extraction and resampling. We named the set of coordinate systems as scaffold coordinate systems set. The proposed coding method has three characteristics: the speed of coding is fast and the computation of coding does not consume too much computational resources; point clouds with different number of points are coded into a fixed-length; coded data of the same object have the highest similarity in different observations. Secondly, based on the proposed coding method, this paper proposes a searching based 3D segmentation algorithm that can segment various preset objects from complex point clouds. Similar to deep learning based methods, the proposed method does not rely on manually design feature extractors. And compared with the traditional methods, this method has no need for large amount of computing resources and doesn’t need large training. Experiments show that the proposed method can effectively segment complex objects without obvious features with good adaptability and robustness.