Research And Implementation Of Video Stitching Technology With Large Visual Angle Of Special Scene

Video mosaic technology is a kind of technology that splices multiple video streams into a single video stream.Video mosaic technology has been applied in many fields,including video surveillance system,sports events,etc..In this paper,a three-way 180-degree panoramic video mosaic system is designed.The system is mainly aimed at the image mosaic of the three-way 180-degree panoramic video in football field background.The system is installed on a fixed platform,the overlap area of the three cameras is about 30%,and the vertical distance between the cameras and the ground is set as 2 meters.Due to the application of the 180-degree panoramic video splicing technique,the video can cover the entire football field and allow users to experience a wider perspective and higher resolution.The specific research content in the paper can be summarized as follows:(1)First of all,the focal length parameters of different kinds of cameras are estimated,the camera parameters are corrected,the distortion is eliminated,and the camera suitable for this system is selected.Furthermore,the homography matrix is studied in this paper.Based on the HomographyNet network which is usually applied to estimate the homography matrix,the network structure is improved and the method of deep learning is used to train the homography matrix of the final model,which is more accurate.(2)Referring to the concept of scale space of the SIFT feature detection algorithm,it can solve the disadvantage of ORB scale invariance,the attributes of the improved ORB algorithm scale invariance are strengthened,but also improves the performance of the whole feature detection algorithm.Experimental results show that the improved ORB algorithm can enhance the feature detection performance of the video mosaic technique.(3)Based on the research and analysis of the existing image deformation methods,three channels of video images are projected onto the cylindrical surface in this paper.In view of the gap problem in image stitching,the common image fusion methods are studied and analyzed in this paper.It is found that multi-resolution fusion can reduce the parallax between cameras to a certain extent and make the transition images look more natural.Therefore,multi-resolution fusion is the most suitable for video mosaic system designed in this paper.(4)Because the speed requirements of video stitching cannot be satisfied on CPU,this paper applies the programming model of CUDA framework to optimize several main time-consuming algorithms,such as feature matching,image projection deformation and multi-resolution fusion.Moreover,the above mentioned time-consuming algorithms are further transplanted to GPU.Lastly,the speed of the three 180-degree panoramic video mosaic system can reach 20 FPS by using the high parallel processing ability of GPU.