A real-time robust human motion tracking system is proposed based on image processing. The proposed system is designed to be robust using three fundamental image processing modules called HeadFinder, Head Classifier and HeadTracker. HeadFinder is a head detection system using principle of "motion circle is a head of human", which leads to a robust detection of heads from images. HeadClassifier is a head discrimination system using low level image data and sophisticated classifier called support vector machine, which leads to be an universal discrimination of head images from non-head ones. HeadTracker generates 3D trajectory of the head which is detected by HeadFlnder and verified by HeadClassifier. Finally, HeadFinder and HeadClassifier are combined dynamically to bring complementary modules ( HeadFinder and HeadClassifier) together to create a new robust and universal bead tracking system.
展开▼
