We propose a novel model-based algorithm which finds the 3D pose of an object from an image by breaking d own the estimation process into two linear porcessing stages, namely the depth recovery and the pose calculation. The depth recovery stage determines the new positions f the model point set in 3D space whereas the pose calculation step is a least-quare estimation of the transformation parameters between the point set formed from the previous stage and the model set. The estimates are iteratively refined until converged. The advantage of using our algorithm is that the computational cost is much reduced. We test our algorithm by applying it to both synthetic as well as real time head tracking problem with satisfactory results.
展开▼