In the streaming data case where we process one look at a time and then determine where to move to next, we'd like to be able to identify a circle and move towards it if we are not certain whether it actually is a circle. How can we provide a level of confidence or uncertainty in our classification of a segment as the circular target?