Research On Panoramic Video Stitching Technology Based On Adaptive Seam Line And Fast Fusion

Video surveillance is one of the important means for people to obtain information.With the continuous development of computer and image technologies,there is a growing demand for a more immersive observation experience,leading to the emergence of panoramic video stitching technology.However,existing panoramic video stitching algorithms have some issues,such as the presence of redundancy and incomplete stitching caused by moving objects in fixed camera positions,as well as the inability to quickly eliminate manual stitching artifacts.To address the issues caused by moving objects during the stitching process,this thesis proposes improvements to an adaptive seam line algorithm.The specific process involves calculating the image mapping relationship to determine the overlapping regions,selecting a pixel within the overlapping area as the reference point for the seam line,and preserving multiple seam lines.Additionally,the algorithm utilizes background frame differencing to determine the mask image generated by the seam line and the moving objects within the overlapping region,and selects an appropriate mask image for stitching to achieve adaptive video stitching effects.By applying this algorithm,the issues of redundancy and incompleteness in video stitching are effectively mitigated.To tackle the problem of quickly eliminating manual stitching artifacts during the stitching process,this thesis proposes improvements to a multi-band fusion algorithm.The process involves utilizing the mask image data obtained from the aforementioned adaptive seam line algorithm,utilizing GPU acceleration,performing operations such as upsampling and downsampling on the mask image and the mapped image using the characteristics of the CDUA kernel function,and simultaneously conducting Laplacian pyramid reconstruction and mapping storage.By applying this algorithm,the processing speed is significantly improved without compromising the stitching quality.To validate the effectiveness of the proposed algorithms,this thesis constructs a real-time panoramic video stitching platform with both hardware and software components.The hardware platform includes Daheng industrial cameras,NVIDIA RTX 3060,3D models,and external triggering circuits.The software platform includes Visual Studio 2017 and the Open CV computer vision library.The platform employs a modular design,consisting of preprocessing,initialization,accelerated processing,and image display/storage modules,and applies the improved algorithms proposed in this thesis for processing.Experimental results demonstrate that the system can perform real-time processing on three video streams with a resolution of1600×900 and output panoramic video frames with a horizontal field of view of 180°.