融合图像语义的动态视频拼接方法

Aiming at the problem that only low-order geometric features of images are considered in the current video stitching methods without consideration of the high-order semantic features of images, we propose a dynamic video stitching method embracing image semantics, which adopts the multi-threaded programming for accelerating stitching speed. The main contributions are an adaptive dynamic video stitching system framework, an efficient feature matching algorithm embracing semantics image, and video stitching quality evaluation algorithm embracing semantics. The vehicle-mounted cameras are applied in the experiment of dynamic video stitching in campus environment for verifying the effectiveness of the proposed algorithm. Obviously, compared with the traditional algorithm, the accuracy rate of new matching algorithm dramatically increases by 50%, the result evaluated by new stitching qulity evaluation algorithm are also more suitable for the human visual sense, and the stitching quality of the proposed stitching method increases by about 25% with remarkable robustness and accuracy.