I recently came across an article of a Japan store using face recognition to identify possible shoplifters. My immediate hunch was that the technique they are using is the study of body pose because a person could tell the difference if a thief is being suspicious by looking at how he behaves. So I believe we can train a neural net to do the exact same thing.
But I could not figure out how to represent this information in a neural net as we need to encode the information of a time unit in order to succinctly determine that a person is suspicious.
Anyone knows any similar project and mind to shed a light?