Online body tracking by a PTZ camera in IP surveillance system

In this paper, an online human body tracking method by an IP PTZ camera based on fuzzy-feature scoringe is proposed. Because the surveillance system uses a built-in web server, the camera control entails camera response time and network delays. Thus, the frame rate is irregular and in general low (3-7 fps). Our method has been designed specifically to perform in such conditions. At every frame, candidate targets are detected by extracting moving target using optical flow, a sampling, and appearance. The target is determined among samples using a fuzzy classifier. Results show that our system has a good target detection precision (> 88%), and the target is almost always localized within 1/4th of the image diagonal from the image center.