Research On Video Stitching Technology For Optimizing Moving Objects

With the development of information technology,people are increasingly demanding on videos.One of the most important requirements is the expansion of video vision.Video stitching technology is to stitch multiple videos,which are simultaneous and have overlapped areas,into one video with a larger field of vision to meet people's demand for information.Up to now,this technology has a wide range of application scenarios.In security,the technology can stitch multiple surveillance videos into a panoramic video which can relieve the pressure of security staff.In medicine,multiple probes often enter into the human body to take pictures at the affected area.Video stitching technology can combine the captured videos into one video,which is help for doctors to observe the affected area and improve the accuracy of the diagnosis.In aerial photography,videos taken by multiple drones are relatively scattered.This technology can make them into a large-view video,which not only presents a shocking effect,but also facilitates people's observation.This paper will propose two video stitching algorithms to optimize the effect of moving objects in overlapped area.Aiming at the problem that homography leads to inaccurate pixel registration in the overlapped area of left and right images,a video stitching algorithm based on optical flow is proposed in this paper.The algorithm is described in detail from the aspects of overlapped area computation,optical flow estimation and clustering,optical flow field propagation,adaptive video stitching and occlusion area correction.In this paper,we leverage the optical flow field in the overlapped area,which provides pixel-wise dense projection,so that the artifacts can be drastically reduced.This method can effectively improve the accuracy of image registration and optimize the blurring around moving objects in overlapped areas.Under the framework of convolutional neural networks,this paper also proposes a new video stitching algorithm by taking advantage of convolutional neural networks in digital image processing.The algorithm implements an end-to-end video stitching network by using VDSR,residual blocks and batch normalization layers.The input of network is left and right images,and the output is panorama.The proposed network is used to make the left and right images more accurately matched,thereby achieving the purpose of optimizing the panoramic images.This algorithm innovatively applies convolution neural network to video stitching technology and also can achieve a good stitching result around moving objects.