针对密集交通场景中的客流检测问题，提出了基于支持向量机（SVM）多目标检测与Mean Shift跟踪相结合的方法。首先采用自适应检测窗口提取梯度方向直方图，经过SVM分类和聚类算法，得到头部图像初始假设。然后采用Mean Shift算法，对头部假设进行跟踪，得到连续的头部图像序列。通过SVM分类器对序列图像进行整体判断，得到客流信息。实验结果表明，自适应滑动窗口的方法减少了特征提取阶段的处理时间，提高了检测速度；同时，通过对得到的跟踪序列进行整体判别，客流量的检测精度得到了提高。

In this paper we proposed a method for determining the passenger flux based on SVM(support vector machine)detection and Mean Shift tracker. With adaptive detection window, the histogram of the gradient orientations is extracted through the detection region. After classification and clustering, the initial head hypotheses are obtained. Then, the Mean Shift tracker is used to track them, and image sequences of the head are achieved. By the whole decision on the consecutive head sequences using SVM, the number of people is obtained. Experimental results show that the adaptive sliding window method reduces the time consumption, and the accuracy of detection is improved by the combination with head tracking.