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

i.e., if a blob is judged static, we increase the TOM value of all pixels of that blob by one; if an object is judged moving, we set the TOM value of the pixels of that object to zero.

If the TOM value of an object is higher than TTM, we add the all pixels of the object to the background samples (an object has remained stationary long enough to now be considered as background).

Figure 4 (c) shows the result obtained by background sample update at both pixel and blob level. We can see the pixels at the center part of the person are correctly marked as foreground pixels.

2.3.6 The Complete Framework of the Proposed Method

Figure 5. Block diagram of the complete framework.

The major components are shown in Figure 5. The proposed framework mainly contains three phases: extracting all possible foreground pixels, running the SACON algorithm, and validating the pixels inside the foreground holes. In the first phase, we use adjacent frame differencing [27] to extract possible candidate foreground pixels. The computational speed is improved by this as typically only a few pixels (from moving foreground objects) will be dealt with in the second phase. However, if the background includes dynamic parts, some of the extracted candidate

16