为了提高公交客流高峰期时的客流检测准确率，提出了一种可以应用于嵌入式平台的基于机器视觉的客流检测方法。该方法以提取乘客的头部轮廓特征作为主要手段，采用针对非标准圆形，即类圆检测的改进型Hough变换，并针对Hough变换结果进行了结合模糊置信度的感知聚类，有效地去除虚假候选头部轮廓，从而实现视场中每个乘客的准确定位。现场实验结果表明，应用该方法进行公交客流统计，准确率可达85%以上。

To improve the accuracy rate of passenger flow estimation during rush hour,a vision based procedure to estimate passenger flow in buses was presented for the embedded application.The contour feature of the passenger’s head was exploited to locate every passenger’s position.In order to eliminate the false candidate head contour effectively and obtain the position of every passenger accurately,the modification of the Hough Transform to detect quasi circle and the perceptual grouping with fuzzy measures were applied.The results of field tests show the accuracy rate may reach above 85 percent by using the proposed method to estimate the passenger flow in bus.