Study On Algorithms Of Video-based Moving Object Extraction

In the rapid development of the science and technology, living standards in modern society is rising, the security of parks, warehouses, supermarkets and other places people gathering is becoming more important. The emergence of intelligent monitoring system makes monitoring easier, include these places where the monitor staff could not take attention to or be neglected, even those places couldn't be monitored, then notify the relevant people illegal activities, it can reduce unnecessary losses. The first step of the intelligent monitoring system is moving object extraction, it separates foreground needs to be analyzed from the corresponding background, it is the critical technology and a very important step, it affects the following work such as behavior analysis.This paper studies how to extract the moving object effectively in complex condition that periodic moving objects exist include rock fans, ventilation fans, pendulum, and it often has a certain light affection. In the places of the periodic moving objects, the changes of the pixel is regular, we use a codebook to store different values for every pixel, and a pixel often needs several code words to save different ranges. When a new image appears, traversing all pixels, if we find a code word matches the pixel value, it is in the background, otherwise, the pixel belongs to foreground. The ground has many changes, such as switching lights or moving objects, we propose a new algorithm that is the combination of running mean and codebook, running mean is to update the codebook, select background pixels for running mean. Firstly, we create a codebook to construct the background, code books of each pixel record different ranges of pixel values. Secondly, select code words belong to background according to a certain percentage. Finally, the codebook of every pixel updates with the update of running mean, running mean likes a counter in the improved codebook, counts for every pixel, if a object is removed or a pixel value changes for a long time, then it belongs to the background, we create a new code word to record the pixel value, added to the codebook, when the pixel value appears again, we can find the new code word from the codebook, then it will be identified as the background pixel. The method eliminates the disturbance of the repetitive movement. In the platform of Visual Studio 2005, we do experiments in various conditions, compare the experiment results of running mean, Gaussian mixture and codebook, we find the improved codebook has better detection results quantitatively in the condition with periodic movements or frequent moving objects, and the extraction of moving objects is more completer, the outline is more clearer, and it is suitable for background changing and real-time processing, the improved codebook detects perfectly indoor, even in the complex background with frequent moving objects, it also detects effectively.