A high-accuracy online calibration method for structured light 3D measurement. (31st March 2023)

• Record Type: Journal Article
• Authors: Xing, Chao; Huang, Junhui; Wang, Zhao; Duan, Qiongqiong; Li, Zijun; Qi, Miaowei
• Is Part Of: Measurement. Volume 210(2023)
• Publication Date: 2023-03-31
• Subjects: Online calibration -- Matching point -- Topological similarity -- Fundamental matrix -- Trinocular epipolar -- Nonlinear optimization; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2023.112488
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: An online calibration method of a structured light 3D measurement system is proposed, which has been implemented automatically and simultaneously during measurement and solves accurate parameters with the roughly same accuracy as traditional offline calibration. The matching point filtration based on the local point set topology similarity RANSAC, which acquires high-accuracy and correctly-matched matching points and outperforms the typical RANSAC to outliers. The fundamental matrix optimization is performed based on trinocular epipolar constraints. The scale factor solution with spatial field constraints is carried out automatically. All calibration parameters are nonlinearly optimized based on the accurate initial values. Abstract: Traditional offline calibration of a structured light 3D measurement system is cumbersome, ineffective, and dependent on calibration targets, whereas state-of-the-art (SoTA) online calibration methods are often haunted by low-precision matching points and thus exhibit poor accuracy and reliability. To this end, we present an automated and high-precision online calibration approach for structured light probes from a multi-camera-projector structured light system(MCPSLS). At the core of the approach, a new criterion for matching points filtration is constructed based on local point set topological similarity, which demonstrates higher accuracy and more robustness than typical RANSAC to outliers. Moreover, the fundamental matrix is optimized based on trinocular epipolar constraints, which further prevents the extrinsic parameter solution from being inaccurate and unreliable due to the small number of filtered matching points. Besides, to achieve a simultaneous and automated online calibration, the MCPSLS's registration objects are served as spatial constraint fields to solve scale factors. Finally, all calibration parameters are nonlinearly optimized for their global optimal solutions. Experimental results show that the proposed approach outperforms SoTA online methods in robustness and accuracy and achieves comparable performance as the traditional offline calibration method.