Adaptive weighted motion averaging with low-rank sparse for robust multi-view registration

Motion averaging has recently been introduced as an effective means to tackle the registration of multi-view range scans. This approach can view parts of pair-wise motions with high reliability as an input to estimate the global motions for a multi-view registration. However, reliable pair-wise motions are not easy to confirm in most practical applications without prior knowledge. In this paper, we propose an adaptive low-rank sparse (LRS) weighted motion averaging method for a robust and accurate multi-view registration, which can directly reconstruct high-quality 3D shape models from a set of unordered range scans. Specifically, we first introduce LRS matrix decomposition to automatically compute the initial global motions. The LRS matrix decomposition can recover the initial global models through the full exploration of a set of pair-wise motions. Subsequently, we extend the motion averaging with an adaptive weight computation by developing an optimization strategy using the Lagrange multiplier method, which can adaptively compute the weights of the reliability for each pair-wise relative motion. Accordingly, the proposed method can recover accurate and robust global motions in a set of iterations through weighted motion averaging. Experimental results on several public datasets demonstrate the excellent performance of the proposed method in comparison with state-of-the-art multi-view registration and 3D scene reconstruction.