Real-time multiple people detection using skin color, motion and appearance information

Recent research in video surveillance has shown the advantages of multiple human detection and tracking. This paper proposes a novel method for detection of multiple people using robust skin color, background subtraction and human upper body appearance information. We extract the human candidate regions using color transform and background subtraction and update. To classify humans and other objects that have similar skin color region or motion, an efficient incorporation of geometric pixel value structure and model based image matching using Hausdorff distance is implemented. The experimental results show that the proposed algorithm can detect humans under various conditions such as skin color noise and complex backgrounds. The extracted human information can be applied to many natural interface or intelligent video surveillance system.