Passenger counting based on Kinect
Jianzhong Xu, Zhu Qiuyu, Yuan Sai, Suo Wenjun · 2014
People counting is a very promising intellectual application in video surveillance system. Presently, a more advanced counting method is based on the three-dimensional visual tracking. However, conventional tracking method has some defects such as: complex three-dimensional visual model, poor accuracy, difficult to match object and so on. In this paper, we adopt Microsoft Kinect and OpenNI driver as the image input source, which can provide scene depth image and color image simultaneously. We carry out target detection based on threshold segmentation, mathematical morphology and perspective projection; and propose a target tracking method based on Bayesian tracking framework. The experimental results show that the method has a high target detection rate and high target tracking counting accuracy.