Edge optimized extraction from the organized point-cloud data base on the gradient clustering

When the traditional 2D edge detectors are applied to detect object edges in low-resolution depth images, the detection accuracy is poor and the recall rate is low. At present, the existing edge extraction methods based on the 3D point-cloud data have poor real-time performance and weak anti-interference ability. To address these issues, an edge optimized extraction method based on the gradient clustering is proposed to fast and stably detect the 3D edges of objects from the organized point-cloud data. First, the flying pixel noise is filtered to eliminate false detection on the edge by analyzing the distance between neighborhood points. Secondly, an edge/no-edge point separation method based on the gradient clustering is proposed to fast extract the rough edges of objects. Finally, the combination of the fast parallel thinning and the mask filtering is employed to optimize the rough edge. In this way, the precise edges are obtained. Experiments are implemented on the public datasets and a dataset collected by a TOF depth camera to evaluate the proposed method. Results show that the proposed method is superior to the existing methods in the real-time and detection accuracy. With the real data, the edge detection is accuracy 89%, and the FPS achieves 28 fps. © 2022, Science Press. All right reserved.