People Tracking by Integrating Observations from Distributed Cameras and Laser Range Scanners

For tracking a varying number of people using distributed cameras, it is important to deal with appearance/disappearance of people due to various causes including occlusions. We propose a method for tracking people with automatic initialization by integrating observations from distributed cameras and laser range scanners. In our method, the problem of estimating 3D positions and orientations of multiple people’s heads is formulated based on a mixture particle filter. Proposal distributions of a mixture particle filter are constructed by using a mixture model that incorporates information from an optical flow and measurements from a laser range scanner. A set of AdaBoost-based classifiers trained to detect a face with a particular orientation are selectively used for hypothesis evaluation. Our experimental results demonstrate the effectiveness and robustness of our method.