A Bayesian-Based Improved Algorithm for Non-Rigid Structure Reconstruction

Non-rigid structure from Motion (NRSfM) is an important research task in computer vision, aiming to recover the motion information of three-dimensional non-rigid structures from a set of two-dimensional images. An algorithm based on the Riemannian manifold has shown promising results in non-rigid 3D reconstruction. However, this method employs fixed parameters for the reconstruction process, which makes it challenging to find the optimal parameter configuration for complex non-rigid objects. So, a novel NRSfM algorithm based on Bayesian parameter optimization is proposed in this paper. By leveraging the Bayesian optimization framework, this approach adaptively adjusts algorithm parameters via sampling the parameter space and evaluating depth errors, enhancing its suitability for diverse non-rigid structure recovery scenarios. A series of experiments and comparisons have been conducted in this study. Through experiments on non-rigid datasets, we have validated the superiority of the Bayesian parameter optimization method in NRSfM. Experimental results show that the proposed Bayesian parameter optimization significantly improves recovery accuracy compared to existing fixed parameter selection methods.