A Research Of Video Stitching Based On The ORB Feature

In recent years, video stitching technology has been widely used. The video stitching application scenarios and features in video surveillance, remote monitoring, virtual reality and other fields. Researchers proposed many excellent video registration algorithm and fusion algorithm. These algorithms are suitable for specific areas. video splicing techniques exist many difficulties in the application, such as changes in diversity, scene complexity, high real-time requirements.Taking into account the large number of frames, high redundancy features and real-time requirements.On the basis of the existing video splicing method,we propose a video stitching method based ORB features. The method is mainly targeted at single-lens moving under scanning captured video images, the method mainly includes three parts:1st. using image registration method based on characteristics of the ORB.,instead of SIFT, SURF and other image registration method which are now widely used. Compared with the above method, the ORB use more efficient feature extraction and formulation methods to reduce the amount of computation and memory requirements, improving the efficiency of image registration, but also for lighting changes affine transformation, scaling scale has good robustness.2nd. We innovatively proposed a fusion method that computes the bilateral linear weighting between the measure of the image shift. It can automatically handle different directions of the transition region into a fade-weighted, effectively reducing the seam appears. And the method of compression of a transition region to reduce the fusion splicing region, appears to reduce the possibility of blur.3rd. Combining the characteristics of aerial imagery and the image vision movement, a new estimate represents the overlap region, combining computing overlap area, select the image of key frame. Keyframes and adjust the sampling step size based on the speed of the adaptive image shift change. Keyframe stitching, reducing redundant video splicing appears. The combination of these approaches devised a more comprehensive method of splicing aerial video, through targeted improvements, the method can also be applied to video splicing in other scenarios.In this paper, our method have done many comparative tests with the current mainstream computing methods. Experimental results show that the proposed method can effectively reduce the amount of calculation, reducing computation time stitching, and stitching of a video. The method herein may work well in the case of having a lot of interference, and achieved good mosaic quality.