Automatic activity classification
Here is an example of the system learning to classify patterns of activity. We have asked the system to take all of the motions observed over a day, and break it up into the most likely classes of motion.
At the first level, it does this by separating out motion based on direction.
At the second level, it further divides based on size of object.
Further levels subdivide based on shape, speed and location of the object.
All of this is learned automatically by the system, with no input from the programmer other than the motion sequences.