Panoramic video stitching in multi-camera surveillance system

This paper concerns the problem of video stitching automatically in a multi-camera surveillance system. Previous approaches have used multiple calibrated cameras for large scale surveillance and monitoring. In this work, we formulate video stitching as a multi-image matching and blending problem, and use SURF to find matches of key points between images, employ homography matrix to calculate overlapping pixels and finally implement boundary resample algorithm to blend images. In addition, camera postures are estimated and fixed automatically according to key points' information. An experimental surveillance system we implemented confirmed the stability and efficiency of our algorithm.