Automatic segmentation and model identification in unordered 3D-point cloud

Segmentation and object recognition in point cloud are of topical interest for computer and machine vision. In this paper, we present a very robust and computationally efficient interactive procedure between segmentation, outlier detection, and model fitting in 3D-point cloud. For an accurate and reliable estimation of the model parameters, we apply the orthogonal distance fitting algorithms for implicit curves and surfaces, which minimize the square sum of the geometric (Euclidean) 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 applications, e.g., robot vision, motion analysis, and coordinate metrology. To achieve a high automation degree of the overall procedures of the segmentation and object recognition in point cloud, we utilize the properties of implicit features. We give an application example of the proposed procedure to a point cloud containing multiple objects taken by a laser radar.