This paper describes a method for building visual “maps” from video data using quantized descriptions of motion. This enables unsupervised classification of scene regions based upon the motion patterns observed within them. ...