Comparison of Background Models for Video Surveillance
Matthew Fettke, Matthew F. Naylor, Karl Sammut, Fangpo He · 2002
Background modelling is a common form of motion detection employed by many autonomous video surveillance systems. Accurately modelling the background is a challenging task, particularly for outdoor scenes where factors such as background motion and camera shake can cause the mistaken detection of foreground objects. Recent research has developed background models that are capable of detecting foreground motion in real-time while ignoring most of the background motion, but it is not clear how well these models would perform on outdoor scenes that exhibit typical video surveillance problems. The aim of this paper is to assess the performance of leading background models (namely , the Hybrid Detection Algorithm, and Three-frame Temporal Difference), using video sequences that contain problems which trouble existing video surveillance systems. The strengths and weaknessesof these background models are reported and analysed, with the aim of identifying suitable directions for the development of robust background models for motion detection in outdoor video surveillance systems.