Separable Spatiotemporal Priors for Convex Reconstruction of Time-Varying 3D Point Clouds

摘要: Reconstructing 3D motion data is highly under-constrained due to several common sources of loss during measurement, such as projection, occlusion, or miscorrespondence. We present a statistical model data, based on the Kronecker structure spatiotemporal covariance natural motion, prior motion. This expressed matrix normal distribution, composed separable and compact row column covariances. relate marginals distribution shape, trajectory, shape-trajectory models art. When marginal shape not available from training we show how placing hierarchical over shapes results in convex MAP solution terms trace-norm. The fit single sequence, outperforms state-of-the-art methods at reconstructing presence significant loss, while providing estimates imputed points.