| The video splicing technology is necessary for extending the limited view aspect in a single frame from one camera. The extended frame could be composited from different frames of multiple cameras or a single dynamic camera. Based on the image mosaic technology, this paper study the video splicing technology by using the video forming principle and the parallel computing technology. As a result, a multi-camera real-time splicing system and a man-machine dynamic video panorama stitching system have been accomplished.The multi-camera real-time stitching system is presented based on image splicing technology. By calculating the possible redundancy of the camera continuous frames, with the condition of the camera fixed in the relative position, our system can reduce the redundancy through matching and stitching algorithms. Finally the real-time dynamic splicing of cameras is realized on GPUs. Based on this stitching system, we also build a video abstraction application, which can reduce the length of one video without losing any dynamic object. The man-machine dynamic video panorama stitching system analyzes the previous works in two cases:moving people and objects with still cameras, or dynamic cameras with still senaries. The system utilizes time-space model algorithm, and refines the video stream in pixel-level depend on the time-space model, finally stitching the video panoramas which include all dynamic objects with dynamic cameras.To conclude, the paper introduces two video splicing methods on multiple cameras and a single dynamic camera respectively. Furthermore, a video abstraction method is also presented based on multi-camera stitching system. |
PDF Full Text Request