Synchronous motion de-aliasing based on optical flow topological sparse weighting

Abstract

This work aims to overcome the limitations of Euclidean distance weighted sparse subspace clustering that does not consider the manifold structure of a rigid body. Here, a sparse subspace clustering method weighted using optical flow trajectory manifold topology was proposed. In the proposed algorithm, the manifold distance of each trajectory in the space-time similarity adjacency matrix was embedded into the weight matrix to solve the sparse coefficient. This ensured that the trajectory with a relatively closed manifold distance became the sparse self-expression dictionary, thereby reducing the clustering aliasing error of synchronous motion. The comparison of experiments between synchronous motion and synchronous swing reveals that the proposed algorithm can reduce aliasing error down to 1%. Finally, the motion segmentation results of Jacquard needle indicate that the algorithm can be potentially used for industrial applications.