Evolutionary Multitasking for Coarse-to-Fine Point Cloud Registration with Chaotic Opposition Search Strategy

Yue Wu, Hangqi Ding, Maoguo Gong, Hao Li, Qiguang Miao, Wenping Ma

Abstract: Point cloud registration is a challenging task in both computer vision and pattern recognition. In general, the success of well-known registration algorithms depends heavily on the assumption of an initial near-optimal transformation. To address this problem, we propose a coarse-to-fine point cloud registration algorithm based on evolutionary multitasking. Specifically, the point cloud registration problem is solved by knowledge sharing before the coarse alignment task and the fine alignment task. In addition, an effective knowledge transfer mechanism and chaotic opposition search strategy are also developed to improve the effective knowledge transfer between tasks and enhance the exploration of more unknown areas in the population, respectively. The performance of the new approach is examined on 14 models from two datasets and compared with 6 competitive methods. Experimental results show that the new approach has the best robustness and accuracy. The new approach improved the registration success rate by 68% through evolutionary multitasking without providing initial transformation information.