面向多种场景的视频对象自动分割算法

摘要

针对当前应用于视频对象分割的图割方法容易在复杂环境、镜头移动、光照不稳定等场景下鲁棒性不佳的问题,提出了结合光流和图割的视频对象分割算法.主要思路是通过分析前景对象的运动信息,得到单帧图像上前景区域的先验知识,从而改善分割结果.论文首先通过光流场采集视频中动作信息,并提取出前景对象先验区域,然后结合前景和背景先验区域建立图割模型,实现前景对象分割.最后为提高算法在不同场景下的鲁棒性,本文改进了传统的测地显著性模型,并基于视频本征的时域平滑性,提出了基于混合高斯模型的动态位置模型优化机制.在两个标准数据集上的实验结果表明,所提算法与当前其他视频对象分割算法相比,降低了分割结果的错误率,有效提高了在多种场景下的鲁棒性.%Aiming at the problems of poor robustness in the complex environment, lens movement and light instability, a video object segmentation algorithm combining optical flow and graph cutting is proposed. The main idea is to improve the segmentation result by analyzing the motion information of the foreground object and obtaining the prior knowledge of the foreground area on the single frame image. Firstly, the motion information in the video is collected by the optical flow field, and the prior knowledge of the foreground object is extracted. Then, the foreground object segmentation is realized by combining the priori areas of foreground and background. Finally, in order to improve the robustness of the algorithm in different scenarios, this paper improves the traditional geodesic saliency model, and employs the dynamic position model optimization mechanism based on Gaussian Mixture Model based on the intrinsic temporary smoothness of video. Experimental results on two benchmark datasets show that the proposed algorithm reduces the error rate of the segmentation results compared with other video object segmentation algorithms, which effectively improves the robustness in many scenarios.