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

It is clear that the value of the threshold Tn should be influenced by the sample size N: the larger N is, the larger Tn should be. Likewise, Tn should also reflect the error tolerance Tr: the larger Tr is, the larger value Tn should be. Thus, Tn can be effectively set to τ Tr N, where τ is a constant and is chosen empirically.

In contrast to the MOG-based background model, parameters such as the number of modes, and the weight, mean, and covariance of each mode are not required. In contrast to the kernel-based background model, SACON is more computationally efficient and no pre-calculated lookup tables are used.

2.3 Building on SACON: A Background Modelling Framework

In this section, using SACON as a core step, we present a complete framework for background subtraction. Each component will be discussed in turn before we present the overall framework.

2.3.1 Shadow Removal and Related Issues

RGB color space is sensitive to changes of illumination. For example, employing RGB color space may cause incorrect labelling of shadows as foreground pixels. Normalized color has been used in many background modelling methods, such as in [3, 20, 21, 22], to minimise the effects of brightness changes. The normalized chromaticity coordinates can be written as:

r=R/(R+G+B)

g=G/(R+G+B)

b=B/(R+G+B) (3)

(Note: we scale r, g, b to the range [0, 255], assuming an 8 bit image value in each channel is used). However, the complete loss of the intensity information can be a problem so we use ( r, g, I) coordinates [3, 21, 22].

8