Modelling of the background (“uninteresting parts of the scene”), and of the foreground, play important roles in the tasks of visual detection and tracking of objects. This paper presents an effective and adaptive background modelling method for detectin

out map at pixel m at frame t. This map is simply incremented at every frame the pixel has been classified as foreground:

TOMt(m)=TOMt 1(m)+1ifBt(m)=0 otherwise TOMt(m)=0 (7)

In other words, TOM is used to record how long (how many frames) a pixel is continuously classified as a foreground pixel. Once the pixel is classified as a background pixel, the TOM value of that pixel is set to zero. When the value of TOM at a pixel is larger than a threshold TTM (which we experimentally set as 45), that pixel will be assigned to the background (the pixel of the object has remained in place too long).

Our TOM is similar to the Detection Support Map (DSM) in [10] as both work as a counter. The differences between TOM and DSM are in that: (a) TOM represents the times a pixel is classified as a foreground pixel while DSM is used to record how long a pixel is classified as a background pixel; (b) when a pixel is classified as a background pixel, the corresponding TOM value is set to zero; In contrast, when a pixel is classified as a foreground pixel, the DSM value at that pixel is

unchanged.

(a) (b) (c)

Figure 4. (a) A frame of MO; Results obtained by updating at pixel level (b) and combination of pixel and blob level(c).

We find the proposed updating method at pixel level works in most cases. However, in some cases, pixels of moving objects can be incorporated to the background. Figure 4(b) shows such an 14