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

efficient in computation.

In the first part of this paper, we propose a robust and efficient background modelling method, SAmple CONsensus (SACON), and we apply it to background subtraction. SACON gathers background samples and computes sample consensus to estimate a statistical model at each pixel. SACON is easy to perform but highly effective in background modelling and subtraction. Quantitive experiments show the advantages of SACON over several other popular methods in background modelling/subtraction. Such background modelling can be useful in its own right but this paper goes on to tackle the problem of tracking/segmenting people through video sequences. Tracking people is one of the most challenging tasks in computer vision. Human motion is non-rigid because when people walk towards or away from the video camera, both the shape and the size of the images of those people change. People can also merge to form a group, occlude each other, or split from each other. A visual tracker needs to cope with such complex interactions. In the second part of this paper, we again use a form of sample consensus, this time to model the appearance of human bodies. We use the obtained appearance model to segment and track people despite occlusions. We exploit both the spatial and color information of the human bodies in our method. We show experimental results in several video sequences to validate the effectiveness of the proposed method.

The main contributions of this paper can be summarized as follows:

We exploit the notion of "sample consensus" to construct an effective and adaptive

background modelling method (SACON) for detecting foreground objects in both static and dynamic scenes;

We present a new sample consensus based method for modelling human appearance and

handling occlusion in human segmentation and tracking tasks.

3