The following videos show person tracking and simple pose and interaction recognition. These were among the first projects completed:
Posture Recognition: The purpose of this project was to demonstrate that a user could theoretically control a robot by providing only visual cues. A robot was programmed to recognize simple postures: standing, arms-up, kneeling (with object), and kneeling (without object). The same vision module also recognizes four different cone patterns and a small box. The first video segment provides posture recognition examples. The second segment shows a a clip from a trial that featured a robot, that had different cone-visiting routines activated afer observing different postures. The final segment shows a person leading a robot down a hallway. The following robot used only visual information to follow the user, while maintaining a specific trailing distance. All processing is done on-line, in real-time.
Interaction Recognition: In this project, a HMM was trained to distinguish between simple interactions that occur between people.