A method tracks an object. A sequence of visible images and a sequence of infrared images are acquired concurrently of a scene. A visible background image is maintained from each current visible image, while an infrared background image is maintained from each current infrared image. Each current visible image and the visible background image, and each current infrared image and the infrared background image are compared to generate corresponding changed pixels. Shadow pixels and reflectance pixels are discarded from the changed pixels to identify foreground pixels associated with an object in the scene.