双目协同动态背景运动分离方法

Moving object extraction algorithm under dynamic background using two-camera collaboration

针对动态背景中运动分离的难点问题,提出了一种双目协同的运动分离方法。该方法利用一个相机实现监控场景的广域监视,另外一个相机实现运动目标的动态捕捉,通过图像配准的方法协同分离出运动前景。针对两相机同步帧图像存在的尺度和灰度差异,设计了一种由粗到精的图像配准方法,该方法比传统的特征配准方法具有更高的配准精度。通过实际监控场景下的视频实验,验证了方法的有效性和可行性。实验结果表明:该方法对于摄像机运动、尺度和灰度差异等影响因素具有较强的适应能力,在智能监控领域具有较广泛的应用前景。

Moving object extraction from the dynamic background is a difficult problem because of the non - sta- tionary background caused by camera motion. To address this problem, an algorithm framework based on two - camera collaboration was presented. The proposed method used one static camera to capture large - view informa- tion thus allowing panoramic surveillance, while the other one was a moving camera to capture local - view infor- mation thus allowing detailed observation. The image registration technology was used to cooperatively extract mov- ing objects from synchronized images of two cameras. Since the resolution discrepancy and intensity inconsistence between two images increased the difficulty of image registration. A practical coarse - to - fine method was de- signed to deal with this problem, which had better accuracy than traditional feature based method. The effective- ness and robustness of proposed algorithm framework was demonstrated experimentally on video sequences captured from real outdoor surveillance environment. Experimental result shows that the proposed method has good adapta bility to camera movement, scale and intensity discrepancy. The proposed framework can be used in intelligent vis- ual surveillance with wide application prospect.