Human Identification and Recognition of Emotional State from Visual Input.

摘要

We have developed a robust framework for analyzing human motion, for the purposes of identifying people from their movements, and recognizing the emotional state of people from their movements. The key point of this framework is the integration of data from multiple visual sources, such as gait, facial expressions, and body movements. In addition, the algorithms for integrating the data should be general enough to allow contributions from nonvisual sources, such as speech. This framework has immediate applications in the areas of surveillance and interrogation. In surveillance, we can detect intruders through people identification. In interrogation, we provide an invaluable backup for human interrogators and psychologists by picking up subtle behavioral cues from human motion that an interrogator might miss. By recognizing these cues, we can offer valuable cues to interrogators. Our system opens the way for the quantitative analysis of human communication in general.