一种动态场景下的视频前景目标分割方法

A Video Foreground Segmentation Method for Dynamic Scenes

视频中运动前景目标的分割是计算机视觉领域的一项关键问题,在视频监控、检索、事件检测等多个方面具有重要应用价值。现有视频前景目标分割技术主要针对静态场景,在动态场景下难以获取良好效果。该文提出一种高斯混合模型与光流残差相结合的前景目标分割方法。该方法使用高斯混合模型建模,提取初步的前景区域;利用光流残差进一步滤除其中动态纹理背景干扰;采用形态学处理获得前景目标。实验显示,与现有方法相比,该方法可更准确地从动态场景中分割出前景目标轮廓。

Video foreground segmentation is one of the key problems in the field of computer vision. It has important value in many applications, such as video surveillance, retrieval and event detection. Traditional video foreground segmentation algorithms are mainly designed for static scene and cannot competent in dynamic scenes. In this article, a novel video foreground segmentation method based on Gaussian mixture model （GMM） and optical flow residual is proposed. Firstly, the preliminary foreground region is estimated by GMM; then, the foreground region with dynamic texture is detected by optical flow residuals and removed;finally, morphology is utilized to refine the estimated foreground. Experimental evaluation shows that the proposed method can obtain more accurate foreground region in dynamic scenes compared with existing methods.