Pairwise registration of TLS point clouds using covariance descriptors and a non-cooperative game

• Published in: ISPRS journal of photogrammetry and remote sensing 2017-12, Vol.134, p.15-29
• Main Authors: Zai, Dawei, Li, Jonathan, Guo, Yulan, Cheng, Ming, Huang, Pengdi, Cao, Xiaofei, Wang, Cheng
• Format: Article
• Language: English
• Subjects: 3D representation; Covariance matrix descriptor; Game theory; Registration; Terrestrial laser scanning (TLS)
• ISSN: 0924-2716; 1872-8235
• DOI: 10.1016/j.isprsjprs.2017.10.001

It is challenging to automatically register TLS point clouds with noise, outliers and varying overlap. In this paper, we propose a new method for pairwise registration of TLS point clouds. We first generate covariance matrix descriptors with an adaptive neighborhood size from point clouds to find candidate correspondences, we then construct a non-cooperative game to isolate mutual compatible correspondences, which are considered as true positives. The method was tested on three models acquired by two different TLS systems. Experimental results demonstrate that our proposed adaptive covariance (ACOV) descriptor is invariant to rigid transformation and robust to noise and varying resolutions. The average registration errors achieved on three models are 0.46 cm, 0.32 cm and 1.73 cm, respectively. The computational times cost on these models are about 288 s, 184 s and 903 s, respectively. Besides, our registration framework using ACOV descriptors and a game theoretic method is superior to the state-of-the-art methods in terms of both registration error and computational time. The experiment on a large outdoor scene further demonstrates the feasibility and effectiveness of our proposed pairwise registration framework.