Convex Relaxation with Log-Determinant Divergence-L1 Regularization for 3D Shape Reconstruction

Investigated is the problem of estimating the 3D shape of an object defined by a set of 3D landmarks with their 2D correspondences in a single image. To solve this problem, we use a dictionary of the basic shape with LDD-L1 regularization, which is the construction of the shape space model. Based on the proposed convex optimization method, 3D human pose reconstruction by shape space model and 3D variable shape model was carried out on the mocap database. To improve accuracy and reduce the number of iterations, we use PSO algorithm to optimize initial value of the key parameter. The experimental results show that the improved algorithm exhibits less iterations but higher accuracy, which can be much helpful in practical applications.