Pose Estimation by Applied Numerical Techniques

Pose estimation using analysis-by-synthesis techniques has been a topic within computer vision for quite a time. These techniques often make use of numerical optimization algorithms required to work in high dimensional work spaces. In this article we apply two algorithms which to the best of our knowledge have never been used in this context. The first algorithm is based on a generalisation of the snake principle known from active contour, while the second makes use of the simulated annealing principle. In this paper these algorithms are applied to the problem of hand pose estimation, known to be a very difficult task. The methods are based on silhouette matching; matching input silhouettes with synthetically generated silhouettes, and furthermore makes use of stereoscopic vision. Keywords— Numerical Optimisation, Pose Estimation, Analysis-by-Synthesis, Optimisation Algorithms, Downhill Simplex, Simulated Annealing, Spider