复杂环境下基于图和条件随机域的运动车辆检测

Moving vehicle detection in complex environments based on graph and conditional random field

针对当前车辆检测方法中存在难以有效消除阴影干扰的缺点,提出了一种能够消除阴影干扰的车辆区域检测方法。该算法首先运用选择性背景更新法进行背景相减,获取感兴趣区域,然后提出基于图的区域分割算法,对感兴趣区域进行再分割。该方法充分考虑了视频图像全局和局部的空间信息,根据分割区域的大小自动自适应地调节对图像局部细节的忽略程度,从而获取局部区域特征较为一致的分割块。最后基于分割过程中所具有的马尔可夫属性,运用条件随机域的方法建立分割后验概率分布,求取最大后验概率确定标号,并对具有相同标号的相邻分割块进行合并。

A vehicle detection method,which can eliminate the effects of shadow interference,was proposed to solve the difficulty to effectively eliminate shadows interference in the current vehicle detection methods.Firstly,this method obtained the interested area by background subtraction using selective background updates method.Then,re-segmentation was done for the interested area via the proposed regional segmentation algorithm based on graph.Global and local space information of the video image was under full consideration,so the algorithm could adaptively and automatically adjust the ignored degree for image local details according to the size of the segmentation area,and got segmentations whose local characteristics were more uniform.Finally,based on the Markov property during video image segmentation,the segmentation posterior probability distribution was constructed by conditional random field method.By calculating the maximum posterior probability,the label was determined,and the adjacent segmentations having same label were merged.