Research And Implementation Of Multi-channel Real-time Video Stitching Technology

The continuous rise and improvement of computer vision technology has led to the increasing progress of video stitching technology.The application of panoramic video stitching is also becoming more and more popular,with broad application prospects in the fields of video surveillance,travel roaming and intelligent navigation.The panoramic video is not only capable of monitoring and framing from a large perspective,but also can solve such problems as object occlusion in a single camera video and messy display from multiple perspectives.Obtaining higher quality and better real-time panoramic video is a research hotspot and challenge in the field of realtime video stitching.By studying panoramic video stitching,a real-time panoramic video stitching system based on multi-channel camera platform is designed in this paper,which ensures the real-time performance of video while achieving better stitching effect.The main research work of the thesis are as follows:Aiming at the inflexibility and cumbersome operation of the traditional camera calibration process,the paper analyzes the distortion model of wide-angle camera lens and designs the automatic calibration method of video stream based on checkerboard.The video stream images are collected for calibration at the same time interval,and the average error of the pixel values after calibration of each image is calculated.The images with larger error are rejected to avoid the calibration error caused by the blurring of the movement and the calibration is re-acquired.The calibration method enables a single person to quickly perform camera calibration,which increases the practicability and simplicity.During the video stitching implementation process,it takes a lot of time in the image registration phase.In order to extract image features quickly and accurately,a combined algorithm based on BRISK+GMS is designed and optimized.Based on the fast feature extraction feature of BRISK algorithm,the feature extraction range is limited to the overlap region.And the grid image method is used to divide the overlap region image to make the feature point distribution more uniform,thus reducing the number of feature points that need to be operated.In the feature matching stage after feature extraction,the GMS algorithm is used to eliminate the mismatching pair,and the matching accuracy is further improved according to the two-way matching strategy.A dual enhancement of feature registration in terms of speed and matching is achieved.The quality of image fusion is reflected in the visual effect of stitching video.In order to make the image fusion more natural and smooth,the paper designes and optimizes a regionalized feather blending weighted fusion algorithm.Before fusion,the local color correction method based on nonlinear mask is used to compensate the exposure of the fusion image.For the purpose of reducing the error in solving overlapping regions,a fusion region is constructed and regionalized.The stitching seam is obtained by GraphCut method,and the image fusion is realized by using feather blending weighted algorithm in different regions to obtain the panoramic image with higher fusion quality.At the end of the thesis,a complete prototype of panoramic video stitching system is designed.The CUDA parallel architecture is introduced to optimize the algorithm by using the system GPU.The feasibility and practicability of the system scheme are verified by experiments.