Research On Key Issues Of Video Stitching And System Implementation

With the rapid development of society,people’s living standards are improving and their security awareness is increasing.Not only can video surveillance be seen everywhere in public places,but more and more people are also installing camera heads at home.However,when observing surveillance,multiple video screens displayed on a single screen are usually cluttered and do not facilitate the collection of information from multiple surveillance videos.Therefore,designing a system that can splice the video is of great importance to all of our productive lives.Unlike image stitching,video stitching needs to consider the temporal continuity of stitching results while ensuring the quality of single-frame image stitching.In video stitching,we often encounter the situation that there are moving objects passing through the overlapping area,which is easy to produce parallax artifacts and temporal inconsistency in the stitching results.In addition,when there is chromatic aberration between video images,there will be obvious transition traces in the stitching results,which reduces the quality of the stitched video.In view of this,this paper conducts an in-depth study of the above key issues and the implementation of video splicing systems,with the following main research content:(1)The optimal stitching line algorithm is improved to address the problem of artifacts in the dynamic foreground during the video stitching process and to cause the stitching results to be inconsistent in time.In order to obtain a more suitable initial stitching line,a central term is constructed based on the traditional stitching line algorithm;in order to keep the moving object intact throughout the video,a foreground term is constructed,and the foreground term information is applied to determine whether the stitching line needs to be updated;in order to keep the shape and position of the stitching line from changing strongly before and after the update resulting in In order that the shape and position of the stitches before and after the update do not change strongly enough to cause a clear inconsistency between frames,a time term is constructed to constrain the change.(2)The color correction algorithm is studied for the color difference problem in stitching.The histogram matching algorithm is improved to address the problem that color blocks and local overexposure tend to appear in the generated results after the histogram matching algorithm is used for color correction of images.A threshold is set to screen out the intervals that are prone to large matching errors,and the removed intervals are curve-fitted,and finally the whole mapping curve is smoothed.(3)The video stitching system is designed and implemented.The system mainly consists of three modules: video stream decoding,image pre-processing and image fusion.The pipeline mechanism is used to maximize the CPU usage,and the data is processed in parallel using GPUs based on CUDA kernel functions to accelerate the operation of the system.