A Background Modeling Algorithm Based on Pixel Frequency

Background subtraction is an important method of motion detection, which is a key to obtain background images. The paper presents an effective and adaptive background modeling algorithm for detecting foreground objects. According to the hypothesis that the background pixel appears in image sequences with maximum frequency, the algorithm is a new background modeling algorithm based on pixel frequency, in which the pixels of the image are classified according their frequency and select the category with the largest number of pixels as the background that is the average value of these pixels in the category. A real time and regularly updating methods are used to adapt the background variations in this algorithm. Experimental results show that the algorithm can draw out the background from the image with some moving foregrounds fast and robustly, which can adapt the gradual and sudden illumination changes. Therefore, the new algorithm can be used in real-time video surveillance system for effectively detecting moving targets.