Object Tracking and Local Appearance Capturing in a Remote Scene Video Surveillance System with Two Cameras

摘要

Local appearance of object is of importance to content analysis, object recognition and forensic authentication. However, existing video surveillance systems are almost incapable of capturing local appearance of object in a remote scene. We present a video surveillance system in dealing with object tracking and local appearance capturing in a remote scene, which consists of one pan&tilt and two cameras with different focuses. One camera has short focus lens for object tracking while the other has long ones for local appearance capturing. Video object can be located via just one manual selection or motion detection, which is switched into a modified kernel-based tracking algorithm absorbing both color value and gradient distribution. Meanwhile, local appearance of object such as face is captured via long focus camera. Both simulated and real-time experiments of the proposed system have achieved promising results.