Real-time Stereo-Image Stitching using GPU-based Belief Propagation

Image stitching or mosaicing is a challenging vision problem, especially when considering aspects like high definition content, real-time, and proper compensation of the parallax error of objects at different distances to the camera system. Today many approaches to image stitching exist, most of them deal with medium resolution images, offline processing, or are restricted to objects at similar distance. Our approach is based on calculation of disparities between corresponding points in stereo images by employing standard belief propagation. Instead of computing depth-maps like in previous approaches, we compute stitch-maps modifying cost functions in the underlying Markov random field model, which makes further projection steps dispensable and thereby overall computation more efficient. Our GPU-based implementation is real-time capable, allowing to stitch high definition images at constant frame rates. We show exemplary results from our system to demonstrate the quality of the merged images.