A Consensus Based Method for Tracking Modelling Background S(17)

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

foreground pixels may belong to the background. This issue will be solved in the next phase. In the second phase, we feed the candidate foreground pixels, their as well as their corresponding background samples to SACON. Also, the pixels whose TOM values are larger than a value (i.e., the pixels are labeled as foreground pixels for a long time) will be sent to SACON. SACON is then run (using equation (4) - modified to single channel as in equation (5) as appropriate). The pixels from the dynamic background scene are suppressed by SACON and output of the second phase is the detected foreground (FG) pixels. In the third phase, we form connected components and where there are holes inside the foreground regions we validate these pixels (see section

2.3.3). After the third phase, we update the background samples and the TOM using the way described in section 2.3.5 and go on the next frame.

3. Experimental Validation of the Background Modelling

In this section we first show that the background method can be competitive with other contemporary techniques. We then investigated the effect of variations in the parameters that the method employs. We also illustrate the computational cost/complexity of the method. Except for the case where we investigate the influence of parameters on the proposed method, all parameters are held fixed for all experiments.

3.1 Demonstration of the Background Modelling Capability

Toyama et. al. [27] benchmarked their algorithm “Wallflower” using a set of image sequences where each sequence presents a different type of difficulty that a practical task may meet. The size of each frame of all the Wallflower image sequences is 160x120 pixels. The sample rate of each image sequence is 4Hz. The performance is evaluated against hand-segmented ground truth. In this section, we will evaluate our background modelling method using these image sequences 17