Recognizing human activities based on head movement trajectories

Human activity recognition from movement-related signals or image sequences is a quite challenging problem in computer vision. Human activities can be decoded from various set of communication channels but it is proved that the head has a highlighted role to emphasize the message that is being communicated. Recognizing activities from head movements can be suitable, because the head has a near constant shape and appearance during the communication. The spatiotemporal segmentation of head movements can be also done by analyzing the trajectories. In this study, we give a general model for description and recognition of head movements. The basic idea has been extended by introducing a human activity database to make better decisions during the recognition. The proposed approach takes into consideration facial regions that encode essential information about head movements. The essence of head movements is extracted from motion history image representation and aligned by dynamic time warping. The efficiency of our system is also demonstrated by the recognition of head-drawn letters.