Xi'an Gongcheng Daxue xuebao (Feb 2022)

Target recognition and pose estimation method under three-dimensional vision

  • WANG Qing,
  • JIA Xiuhai,
  • YE Minglu,
  • WANG Qiyu,
  • SHENG Xiaochao

DOI
https://doi.org/10.13338/j.issn.1674-649x.2022.01.012
Journal volume & issue
Vol. 36, no. 1
pp. 85 – 93

Abstract

Read online

In order to improve the recognition rate of the 3D target point cloud in the occluded environment, we proposed a feature fusion recognition (CV-SHOT) algorithm combining the clustering viewpoint feature histogram (CVFH) algorithm and the signature of histograms of orienTations (SHOT) algorithm. The CVFH feature was used to quickly and roughly recognize the segmented scene point cloud. So as to extract the SHOT feature of similar target point cloud and obtain the model-scene corresponding point set. The 3D Hough voting mechanism was introduced to accurately identify the target in the scene and obtain the initial pose of the point cloud target. Based on the iterative closes point (ICP) algorithm, precise target positioning and pose estimation were achieved, follonved by setting up an experimental environment to test single-object scenes and partially occluded multi-object scenes. The results show that the CV-SHOT algorithm is robuster with a higher recognition rate of over 90%, significantly higher than that of the traditional algorithm. It can effectively recognize and estimate the pose of targets in indoor complex scenes.

Keywords