“…Recent years, research on pedestrian detection and counting has made great progress: [1,2] use regression techniques to learn a map between features and the number of people in the training set and then use the map to estimate the number of people in novel test images or videos; In [3,4], multi-scale windows slide over the whole image and a binary classifier is adopted to determine whether there is a people within the window; literature [5] put forward a method based on facial feature description and SVM (support vector machine) to count pedestrians; literature [6] adopts background subtraction based on threshold to extract object information, then uses connected component detection algorithm, setting the object feature and shape judgment condition and marking object region, finally count the number of people, but it can't remove the influences by some problem such as illumination, rapidly changing weather conditions, people head which are covered completely etc. ; literature [7] construct a novel system that uses vertical Kinect sensor for people counting which equals to find the suitable local minimum regions, then propose a novel unsupervised water filling method that can find these regions with the property of robustness, but it can't handle the situation where some moving object is closer to the sensor than head; literature [8] uses Bayesian Gaussian process to learn the map between holistic features and the number of people. In [9], KLT tracker and agglomerative clustering were used.…”