动态背景下基于光流场分析的运动目标检测算法

针对现有动态背景下运动目标检测算法的不足,提出一种基于光流场分析的运动目标检测算法.首先根据前背景在光流梯度幅值和光流矢量方向上的差异确定目标的大致边界,然后通过点在多边形内部原理获得边界内部的稀疏像素点,最后以超像素为节点,利用混合高斯模型拟合的表观信息和超像素的时空邻域关系构建马尔可夫随机场模型的能量函数,并通过使目标函数能量最小化得到最终的运动目标检测结果.该算法不需要任何先验假设,能够同时处理动态背景和静态背景两种情况.多组实验结果表明,本文算法在检测的准确性和处理速度上均优于现有算法.

Moving object detection based on optical flow field analysis in dynamic scenes

To overcome the limitation of existing algorithms for detecting moving objects from the dynamic scenes, a foreground detection algorithm based on optical flow field analysis is proposed. Firstly, the object boundary information is determined by detecting the differences in optical flow gradient magnitude and optical flow vector direction between foreground and background. Then, the pixels inside the objects are obtained based on the point-in-polygon problem from computational geometry. Finally, the superpixels per frame are acquired by over-segmenting method. And taking the superpixels as nodes, the Markov Random field model is built, in which the appearance information fitted by Gaussian Mixture Model is combined with spatiotemporal constraints of each superpixel. The final foreground detection result is obtained by finding the minimum value of the energy function. The proposed algorithm does not need any priori assumptions, and can effectively realize the moving object detection in dynamic and stationary background. The experimental results show that the proposed algorithm is superior to the existing state-of-the-art algorithms in the detection accuracy, robustness and time consuming.