Moving Object Real-time Detection and Tracking Method Based on Improved Gaussian Mixture Model

摘要

In order to improve the reliability of moving objects detection and tracking, this paper presents a method for moving object real-time detection and tracking based on Vibe and Gaussian mixture model (GMM). This method uses the 'Virtual' background model that is trained by video sequence instead of the first frame image for background modeling. And then the foreground object is extracted based on the pixel classification. Finally, according to the morphological method, the clearer moving targets are conducted to realize the real-time detection and tracking. The experimental results show that, in comparison with the current mainstream background subtraction techniques, our approach effectively works on a wide range of complex scenarios, with faster detection speed and more reliable detection results.