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

however, we cannot simply use some hole filling technique to remove the holes, because these holes may also be caused by the structure of the foreground object or the posture of a human being, for example. Thus, we use a validation procedure to recheck the pixels inside the holes (note: foreground objects are represented by connected components). For pixels inside the holes, we use xtI xbI≤TI (whereTIis a special threshold, we experimentally set it to 7, for the intensity channel applied only to pixels of the holes, as a post-processing step, i.e., if the condition is not satisfied, we mark the pixels of the holes as foreground pixels; otherwise, we mark the pixels as

background pixels).

(a) (b) (c)

Figure 2. An example showing foreground hole pixel validation. One frame of “Camouflage” (a); The results without (b) and with (c) the validation procedure.

Although the validation cannot correct wrong labels of foreground pixels when the color of these pixels is very similar to the background, it can improve the results obtained by Equation (4). Figure 2 shows us an example. One frame of the image sequence “Camouflage” (C) in the Wallflower dataset is shown in Figure 2 (a). The person walked into the room and stood in front of a monitor which has similar color (on the screen) to the person’s clothes. Figure 2 (b) shows the result without the validation procedure. We can see there are a number of holes inside the foreground object, which are wrongly marked as background pixels. Figure 2 (c) shows the result after applying the validation procedure. From Figure 2 (c) we can see that most pixels inside the holes are correctly marked as foreground pixels.

11