基于组合前景提取的混合交通两轮车辆视频检测

针对混合交通流中两轮车辆视频检测问题，提出一种基于混合高斯模型（GMM）与背景累加模型（BAM）的组合前景提取方法，该方法将GMM与BAM组合得到的2 种前景图像分别经过滤波和形态学的膨胀操作处理，然后进行“与”操作，过滤掉高斯前景中的大量噪声，提取出感兴趣前景区域.针对两轮车辆的轮廓边缘特征，采用Canny 边缘检测提取边缘信息，去除前景区域中的非目标区域，采用两轮车辆的自建模板，通过欧氏距离进行模板匹配，定位并标记感兴趣区域中的目标区域.在OpenCV 和Matlab7.1 实验测试平台上，对典型城市混合交通路段的交通流视频进行测试.结果表明，该方法对混合交通流中两轮车辆的识别检测具有较高的准确率.

Video-based Two-wheel Vehicle Detection for Mixed Traffic Flow Based on Combination Foreground Extraction

A two-wheel vehicle detection method for mixed traffic flow based on combination foreground extraction method is proposed. The foreground images extracted by Gausssian Mixture Model (GMM) and Background Accumulate Model (BAM) are carried out with“and”operation after the operations of filtration and expansion of morphology, which can filter out a lot of noise in the GMM foreground and extract the interested foreground fields. According to the contour edge character of two-wheel vehicles, the Canny edge detection method is used to extract the edge information. The self-built template and Euclidean distance are used in pattern matching in order to locate and mark the target area. The experiments are performed using the traffic flow video in the classical urban mixed traffic road. The results show that this method has high accuracy of two-wheel vehicle detection.