Publication

Abstract : Real-time stable panoramic video stitching is a challenging research problem. Reducing computational complexity of stitching algorithm, increasing the quality of the stitched video and a stable repeatable technique for stitching are the major challenges addressed in this paper. We present two major techniques: iterative compositing and pipelined execution. In a scenario where cameras are static, this algorithm dynamically measures the redundancies in the feature detection, feature matching and transformation model estimation stages and reuses the resultant transformation matrices for stitching the subsequent frames. In our experiments, we have determined that our unique algorithm is able to improve the real-time stitching of standard definition videos by a factor of 4.5. Moreover, the quality of the video was found to be satisfactory, devoid of any visible seams, very low jitter and no other visual artifacts.