Moving object detection is a challenging task in many computer vision applications. In this paper, we propose a robust background modelling method for this task. First, the background is updated by an adaptive strategy based on Centre-symmetric Local Binary Patterns. Then, background subtraction is used for detecting moving object. Although the traditional background subtraction technique uses the difference value between the current pixel and its corresponding background pixel for objection detection, our method utilises the confidence factor to determine whether the current pixel is a background or foreground pixel. The confidence factor of the current pixel is calculated in term of the difference values of its neighbourhood pixels. In the experiments, the proposed algorithm is tested on several challenging datasets such as PETS 2009, BMC 2012 and SABS. The experimental results demonstrate that our algorithm can robustly detect moving object under various scenes.
IntroductionMoving object detection, which is also called moving object segmentation, is an intelligent analysis technology to extract moving object from surveillance video sequences. It is a fundamental problem in many computer vision applications, such as video surveillance, industrial automation and intelligence transportation. 1 Although many methods have been proposed for moving object detection, it is still a challenging problem because of dynamic scenes, such as illumination changes, dynamic changes relating to swaying trees, rippling water, fleeting cloud and so on.Generally, moving object detection method can be classified into three categories, which are the inter-frame difference method, the optical flow method and the background subtraction mehod. 2,3 Compared with the inter-frame difference and the optical flow method, background subtraction is a widely used method in practical applications because of its simple realisation, fast computing speed and non-prior knowledge. The performance of background subtraction mainly depends on the background modelling technique. A robust background modelling algorithm needs to overcome background changes, which occur in outdoor scenes, such as illumination changes caused by sun moving and dynamic changes including swaying branches or water fluctuations. Therefore, how to build and maintain an adaptive background model is a key problem of this method and is still a hot research topic. [4][5][6] In the past two decades, a large number of different background modelling methods have been proposed. Due to their pervasiveness in many applications, there is no unique classifying method on the existing works of background subtraction techniques. According to the model parameters, these methods can be divided into methods with or without parametric. In terms of the features used in modelling, there are single characteristic models and hybrid characteristic models. On the basis of pixel for modelling background, background model can be classified into pixel-based model 7-12 and region-based model. [13][14][15][16]...