REAL TIME VIDEO FOREGROUND EXTRACTION BASED ON CONTEXT-AWARE BACKGROUND SUBTRACTION*
Álvaro García, Jesús Bescós · 2004
This paper describes a real-time approach for foreground segmentation in image sequences taken from a stationary camera, as a first preprocessing step for object extraction. The proposed method combines a background subtraction method with the extensible inclusion of context information. It currently considers bidirectional temporal change detection as the a priori context data (i.e., desired objects are moving). This combination has proven to improve the detection of the object boundaries, hence enhancing the reliability of the segmentation masks, with little extra computational load. The tested implementation of the proposed foreground detector shows a good trade-off between efficiency, segmentation accuracy and robustness, particularly compared with some other state-of-the-art methods. Index Terms — video analysis, background subtraction, change detection, context information. 1.