Smart Camera Surveillance System Case Study

The described framework in this study covers 4 main building blocks used in a smart camera surveillance system, which can be categorized as object detection, object recognition, tracking and behavior/activity analysis. This chapter describes the related work and current studies in the literature on each of these building blocks. Meanwhile, some brief information related Xbox Kinect in obtaining the depth of information from the video. A typical configuration of smart video processing framework is shown in Figure 2.1.
Each application has a different interests or needs from smart camera surveillance system, hence requires different method. However, there have one common thing in the process regardless their application fields, which is moving
However, it fails to detect whole relevant pixels of some types of moving objects. Figure 2.3 shows an inaccurate motion detection example for the salient motion detection. Besides, it also fails to detect stopped objects in the frame. In order to overcome this problem, additional methods need to be adopted.
The Gaussian Mixture Model (GMM) is a single extension of the Gaussian model. The single Gaussian models each pixels in a scene with an estimated mean intensity value. However, the limitation of the model is that it only can handle unimodal distributions. Therefore, the GMM is proposed to solve the drawback of single Gaussian model by calculating the probability intensity level function [4]. In the approach of GMM, each pixel in the image is labelled by a mixture of K Gaussian distributions that related with the changes of state from single frame to another. The GMM algorithm is worked with every single frame pixel and then change the colour image to the binary image which are 0(black) assigned to no state changes and 1(white) assigned to drastic changes in state. The frame pixels are eliminated from the video sequence in order to obtain the desired result. Figure 2.3 shows all the moving objects in the video are located using GMM. The GMM is more robust and reliable compared with single Gaussian model or other background method due to the changes in environment/scene. This method is becoming more popular due to its reliability in the scenes such as noise, illumination changes and shadow [49]. However, the drawback of this method is the foreground objects bending too fast into the background. Hence, the problem with slow-moving objects is not recommended by using this