A Real-time Object Detection System Using Selected Principal Components

The detection of moving objects is a basic and necessary preprocessing step in many applications such as object recognition, context awareness, and intelligent visual surveillance. Among these applications, object detection for context awareness impacts the efficiency of the entire system and it requires rapid detection of accurate shape information, a challenge specially when a complicated background or a background change occurs. In this paper, we propose a method for detecting a moving object rapidly and accurately in real time when changes in the background and lighting occur. First, training data collected from a background image are linearly transformed using principal component analysis (PCA). Second, an eigen-background is organized from selected principal components with excellent ability to discriminate between object and background. Finally, an object is detected by convoluting the eigenvector organized in the previous step with an input image, the result of which is the input value used on an EM algorithm. An image sequence that includes various moving objects at the same time is organized and used as training data to realize a system that can adapt to changes in lighting and background. Test results show that the proposed method is robust to these changes, as well as to the partial movement of objects.