Person Detection & Tracking
This module is centered around three main components: 'Stereo Input and Pre-Processing', 'Stereo Multi-Target Detection and Tracking', and 'Visualization'. All of the useful work is done in the first two of these modules, 'Visualization' is only implemented for testing purposes and does not influence the operation of the system.
Our detection algorithm is primarily focused on searching for users' head in photographs. Image intensity and edge detection are used to identify the outline of bodies while image color is used to identify patches of skin. In each case, the head is then separated from the body and an ellipse is fitted to any areas resembling a human head. The areas identified as corresponding to a head in each of these three cases are compared, identifying the objects most likely to be people.
Again, the symbolic results from the detection algorithm are fused together at the symbolic level from the two camera inputs, yielding a highly reliable way to identify and maintain state regarding potential users within the environment. The following picture provides an example of this tracking system in action.
