Gesture Recognition
When STEVI is within close proximity to a potential user, it needs to not only track that user's position within the environment but also detect the orientation of that person's limbs. This allows a user to issue STEVI commands simply by gesturing, at which point STEVI will acknowledge and execute that command.
This is accomplished by representing the human body as a tree structure, beginning with the head. Edge detection and image color are combined with a prediction as to where the subject's limbs should be. Since the location of the subject's head is already known from the head detection module, the search space is considerably reduced. Additional constraints are added by overlaying a model of the human body and computing possible positions through transformations.
As before, further refinements are made by fusing symbolic data from the separate stereo streams available.