Research On The Key Technology Of Multi-view Video Stitching

Multi-view video stitching is widely used in many fields,such as video surveillance,video conferencing and satellite remote sensing.It can be used to expand field of view while ensuring high resolution.The technology has been a research hotspot.The key of multi-view video stitching technology is video frame stitching.However,there are some problems in the current video frame stitching,such as the speed problem of video frame registration,the ghost problem of video frame fusion and the distortion problem of video frame transform.Aim to overcome these limitations,the following four studies were performed in this paper.(1)Video frame registration method based on iterative hashImage feature matching is the key step during image registration.Forced matching and approximate nearest neighbor matching are most common used matching methods.Forced matching can not meet the real-time requirement of video stitching due to its highly computational complexity.kd-tree is usually used for nearest neighbor matching,however,the dimension disaster problem will appear when the data point dimension is high.Therefore,we propose an iterative hash algorithm,which is an enhanced local sensitive hash algorithm.The algorithm can speed up the image registration and improve the matching accuracy.Applying the algorithm to image registration is able to achieve better performance.(2)Video frame fusion method based on dynamic seam-line and local weighting.Weighted average method,multi-band method and deducting foreground map method are frequently used image fusion methods.However,these methods are not suitable for video stitching.Weighted average method will cause ghosting problems when the dynamic object appears,multi-band method cannot meet the real-time requirements of video stitching,deducting foreground map method requires high scene requirements.Therefore,we proposed the video frame fusion method based on dynamic seam-line and local weighting.The best seam can be found dynamically and the seam neighborhood can be fused locally by the proposed method,which can solve the problem of ghosting and ensure real-time video stitching.(3)Video frame fusion method based on shape-preserving half-projective transformation modelIn the case of large camera parallax,the single projection transformation model will lead to a large distortion of video frame image.In order to solve the problem,we apply the shape-preserving half-projective model to video stitching.The transform model utilizes the properties of projection transformation and similarity transformation,which can ensure that the video frame does not deform seriously.The robustness of video stitching are enhanced using this method.(4)Design and implementation of multi-view video stitching systemBased on the previous achievements,we designed and implemented a multi-view video stitching system.The system consists of two components: the video capturing component and the video stitching component.The video capturing component is responsible for designing the interface and displaying the video in real-time.The video stitching component is responsible for stitching the video frame image.