Authors: Tianfei Chen; Yang Wang; Defeng Wu; Xiang Wu; Zi Ma
Addresses: Department of Electrical Engineering, Henan University of Technology, Zhengzhou, 450001, China ' Automation Research Center, Dalian Maritime University, Dalian, 116026, China ' Marine Engineering Institute, Jimei University, Xiamen, 361021, China ' Department of Electrical Engineering, Henan University of Technology, Zhengzhou, 450001, China ' Automation Research Center, Dalian Maritime University, Dalian, 116026, China
Abstract: Lens distortion correction is crucial to obtain the best performance for vision system where rigorous accuracy is demanded. This paper presents a robust non-metric correction method which relies on the fact that straight lines in the real world must always project to straight lines in the image under the pinhole camera model. Neither calibration objects nor any known structure are required. Two types of distortion measurement are defined as objective function, and then optimisation algorithms are adopted to find distortion parameters. The comparison is analysed between the traditional optimisation algorithm, standard particle swarm optimisation (PSO) algorithm and two other modified particle swarm algorithms based on different typical strategies. Experiment results show that the opposition learning-based particle swarm algorithm (OPSO) with a mutation operator has strong robustness, and in the case of low-level noise, the performance of algorithms based on the particle swarm optimisation is better, and the effectiveness of the method is verified by different degrees of lens distortion.
Keywords: computer vision; distortion correction; distortion measurement; particle swarm optimisation; PSO; lens distortion.
International Journal of Modelling, Identification and Control, 2014 Vol.21 No.3, pp.330 - 337
Received: 08 May 2021
Accepted: 12 May 2021
Published online: 27 Apr 2014 *