基于改进3D马尔可夫模型的动态车辆检测

Dynamic identification of vehicles using improved 3D-Markov model

针对复杂的交通视频,提出对车辆进行动态检测.综合利用时间和空间信息,对传统的马尔可夫概率图模型方法进行了改进.通过构建标记场为3D马尔可夫模型、特征场为混合高斯模型(GMM)的混合概率模型,将每一帧的每一像素与它相邻的像素和前后帧的像素相关联,并对模型能量函数进行改进,把像素之间的差值作为能量函数值的标准,实现了边缘提取和运动检测的结合.最后通过贝叶斯估计和迭代条件模式(ICM),对车辆进行动态检测.实验结果表明:此方法可以在复杂的交通视频中准确定位正在运动的车辆,显示运动车辆的轮廓,并将速度转化为输出图像的灰度值,有很好的抗干扰效果.

Method for dynamic identification of vehicles in complex traffic environment was proposed by utilizing temporal and spatial information. The method took into account the connections between adjacent pixels and frames by constructing a probabilistic model whose label field was 3D-Markov model and characteristic field was Gaussian mixture model. Combination of edge extraction and motion detection was achieved with an improved energy function whose standard was the difference between pixels. Finally, the motion detection of the vehicle was performed with Bayesian estimation and itera- tive conditional mode. Experiment results show that the proposed method can extract moving vehicles from complex traffic video by displaying their edge, and the velocity of the vehicle is converted as gray scale value of the output image. The method was also proved to be robust under interferences.