Article Preview
TopOverview Of Background Subtraction
Background subtraction (BS) (Elgammal et al., 2000)(Maddalena et al., 2008)(Haritaoglu et al., 2000)(Stauffer & Grimson, 1999)(Stauffer & Grimson, 2000) is a widely used solution for detecting moving objects accurately. The main idea is to develop a background model for representing the background image that is then updated periodically. The motion can be detected by using multiple frames consecutively (Devi et al., 2016). The Background Subtraction process starts with frame extraction from an input video sequence. The background modeling is done by employing the difference in pixels of the present frame and the initial frame. Using this difference and a set threshold value the foreground is detected. This foreground object detection is not necessarily a simple object motion detection issue. The background may as well contain moving objects such as running water, moving leaves of trees, etc. Simultaneously, the region or object of interest, such as a human, maybe still for some time but needs to be detected continuously. Another set of errors may occur when a static entity, e.g., a parked car, starts to move. In this case, both the moving entity and the empty space, which is left as a difference of frames (referred to as a ghost), are detected as foreground. The process of BS has been shown in Figure 1.
Figure 1. Background subtraction process
This approach works suitably only when static cameras are present. The difference only outputs new objects or the objects that are non-stationary. The technique is very simple and works with affordable computations giving results in real-time. The drawbacks associated with this technique are that these algorithms are exceptionally sensitive to illumination changes, dynamically moving backgrounds and an occurring camouflage in background and foreground objects. Thus, a good background maintenance model is a necessity. The process of background subtraction can be seen with the help of an example, as shown in Figure 2 (Example of Background Subtraction Methods, n.d.).