Research And Implementation Of Real-time Video Panarama Stitching Algorithm

With the acceleration of urbanization,urban environment now is becoming more and more complex.There are large numbers of surveillance videos need to deal with every day.If we stitch neighboring video by common area,it can not only make surveillance video display more convenient and effective,but also reduce the cost of video storage and network transmission.Existing work mainly concentrate on image stitching and there is not enough research on video stitching.This paper proposes a real-time video stitching method based on the relationship between frames.For the first frame of every video,we use traditional method to extract image feature.Then match the feature and solve the image transformation matrix.For other frames,we compute the inter-frame relational matrix and use the matrix obtained from the last step to get the current transformation matrix.In this process,we introduce GPU parallel technology to speed up the feature extraction part and make some optimization to the RANSAC algorithm.Finally,we use multi-resolution fusion algorithm to achieve the seamless image stitching.During the video splicing process,each frame we get has clear picture quality.Experimental results show that our algorithm can complete stitching tasks effectively.