Geometric Segmentation and Object Recognition in Unordered and Incomplete Point Cloud

In applications of optical 3D-measurement techniques segmentation and outlier elimination in point clouds is a tedious and time-consuming task. In this paper, we present a very robust and efficient procedure of segmentation, outlier elimination, and model fitting in point clouds. For an accurate and reliable estimation of the model parameters, we apply orthogonal distance fitting (ODF) algorithms that minimize the square sum of the geometric error distances. The model parameters are grouped and simultaneously estimated in terms of form, position, and rotation parameters, hence providing a very advantageous algorithmic feature for segmentation and object recognition. We give an application example for the proposed procedure which is applied to an unordered and incomplete point cloud containing multiple objects taken by laser radar.