People tracking and following using multiple RGB-D cameras

Being able to locate a specified people and to simultaneously follow him in an unknown environment has become an essential capability of service robots. Lots of efficient methods for implementing this so-called people tracking and following capability have been proposed. However, these mature methods cannot be applied for civil use readily for their high cost. The appearance of low-cost RGB-D camera solves this problem and there have been sorts of methods describing how to build a people tracking and following module with RGB-D camera. For the sake of cost control, the modules based on low-cost RGB-D camera always suffer from noise, low-range and low-precision problem. We present a method that implements people tracking and following based on RGB-D data generated by Microsoft Kinect sensor, in indoor environment. Furthermore, we also discuss a set of methods to optimize the result in order to filter the noise, extend the detected range and decreasing the effect of low-precision.