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

2.3.4 Setting the Value of Tr

(a) (b) (c)

Figure 3. (a) One frame of the videos; the detected foreground pixels by (b) setting a global value to Tr and by (c) setting the values of Tri for each pixel according to Equation (6).

There are two possible ways to set the value of Tr. The first is to empirically set a global value of Tr for all pixels. However, to obtain an effective value of Tr for all image pixels is hard. The second way is to estimate the standard variance σi at each image pixel and set Tr equal to ησi, (where η is usually set as 2.5 or 3). However, σi may be overestimated when the data is multi-modal distributed.

We set Tri for each image pixel by combining the above two ways as follows:

Tri=min(T1,ησi) (6)

where T1 is a constant (We will discuss the influence of T1 on the results in subsection 3.2.1). From Figure 3, we can see that when we set a global Tr, some parts (e.g., part of the trousers of the person in the first row and the shirt of the person in the second row) of the foreground objects are not successfully detected. In contrast, when we set the various values of Tri for each pixel according to Equation (6), most of the foreground pixels are correctly detected (Figure 3c). 12