Dense Stereo Matching with Application to Augmented Reality

This work presents two techniques for computing dense disparity maps from two or more images. These methods are exploited in an application of augmented reality in order to add a virtual object with proper occlusions in the real scene. The proposed stereo matching techniques are based on area matching. We first present an implementation of the dynamic programming which produces a high quality of dense disparity map. Furthermore, in order to improve both the time computing and the map quality, we propose a hierarchical approach which combines the multi-resolution and the dynamic programming. The disparity maps thus obtained are applied in augmented reality in order to integrate in a realistic way the virtual objects. The applicability of the method is shown on many sequences of images.