Three Ways to Improve the Performance of Real-Life Camera-Based Fall Detection Systems

Figure 2

Overview of implementation of foreground (FG) segmentation based on robust background (BG) subtraction using a particle filter (PF) with three different measurement coefficients: FG, histogram and person detection coefficient. (a) BG model. (b) Current input frame. (c) Detected FG. (d) Determined region of interest (ROI) as biggest foreground object (crosses indicate center and top of bounding ellipse (BE)). (e) Histogram coefficient used by PF. (f) FG coefficient used by PF (OL: first outer layer; OOL: second outer layer). (g) Person detection coefficient that controls weight-factor (small rectangle represents detected upper body; large rectangle represents extrapolated person). (h) Prediction of PF used to update BG model selectively slow inside prediction, fast outside.