ViCoMo: visual context modeling for scene understanding in video surveillance