视频监控中的图像拼接与合成算法研究

Image mosaicing and compositing algorithm for video surveillance

视频监控常用云台摄像机或多个静止的摄像机监控视场范围比较大的区域,分别研究了多摄像机的视频合成算法和云台摄像机的视频拼接算法。对于多个摄像机的情况,采用高斯混合模型对各个视频序列进行背景建模。然后根据背景图像计算各个摄像机的单应矩阵,并将多个视频序列合成为一个大的视频序列。对于云台摄像机跟随拍摄的情况,提出了一种基于关键帧的视频拼接算法。根据重叠区域的大小和纹理信息的丰富程度选取关键帧,并把各帧图像对准到其前面最近的关键帧,同时更新重叠区域内的背景模型参数。由于关键帧之间具有比较高的对准精度,因而该方法能够较好地重建运动目标所经过的场景的背景图像。

Multiple static cameras or PTZ （Pan-Tih-Zoom） cameras are often used to monitor activities over a wide area in video surveillance system. This paper studies the video compositing algorithm of multiple static cameras and video mosaieing algorithm of PTZ camera. For multiple static cameras, it uses Gaussian mixtures to model the background of each video. Then it computes the homography of each camera based on the background images, and a large video sequence is composited from the multiple video sources. Moreover, the algorithm also proposes a key frame based video mosaicing approach for PTZ cameras. The key frames are selected based on the amount of overlap and abundance of the texture information. All frames are matched to their latest neighbor key frames, and then background model parameters are updated for the overlapping regions. Key frames have very high alignment accuracy, so the proposed method can create accurate background image of the scenes that the moving object has traversed.