A Traffic Video Background Extraction Algorithm Based On Image Content Sensitivity

Image segmentation is a key procedure from image processing to image analysis, it is also a basic computer visual technology. That's all because image segmentation, target separation, feature extraction and parameter measurement will transformed the original image into more abstract, more compact form, and make the higher analysis and understanding possible. The quality of image segmentation and the accuracy of regional boundary localization will directly affect the following regional description, so the image segmentation has been obtained people's great attention.In the traffic video, one of the most important application of image segmentation is separate the moving objects from background. In order to obtain more accurate target, which you are interesting, the timely background updating is necessary before the moving objects detection, especially in the image difference method. In recent years, the scholars home and abroad have done extensive research on background updating. Common methods of the background update including: Statistical background model, Gaussian mixture model and the background model based on Kalman filter, etc. Statistical background model is applicable in this case, a small number of targets, continuous motion and the background is visible in most of the time. But if a large number of moving objects in the scene, background which estimated by this method will have a big error. Gaussian mixture background model can well adapt to light changes, slow-moving targets, etc. The application of the background model based on Kalman filter is constrained due to without considering that the system noise is changed along with the illumination condition's changes.In this paper, on the basis of the Background Updating Algorithm based on the moving region, presents a Traffic Video Background Extraction Algorithm based on Image Content Sensitivity. Multi-frame images have different Entropy Energy (EE), it uses this feature to analyze image's content. Firstly, obtain the initial background image which has the least Entropy Energy in the moving region through mixture Gaussian background modeling algorithm. Then weight factor is selected dynamically by Entropy Energy and the mixture Gaussian model of every pixel in the current image is updated. Finally, every pixel's value in the background image is updated by weighted average. Experiment results show that this method can well delayed the occurrence time of the stationary vehicles, simple and robust. Especially for the condition that a number of vehicles into or out of the scene quickly, the effect is better.