Résumé: Identifying any moving object is essential for wide-area surveillance systems and security applications. In this paper, we present a moving object detection method based on background modeling and subtraction. Background modeling-based methods describe a model with features such as color and textures to represent the background. Background subtraction is challenging due to complex background types in natural environments. Many methods suffer from numerous false detections in real applications. In this study, we create a background model with each pixel’s age, mean, and variance. Our main contribution is to propose a tracking approach in background subtraction and use simple frame difference to set weight during background subtraction operation. The proposed tracking strategy aims to use spatio-temporal features in foreground mask decision. The tracking method is used as a verification mechanism for candidate moving object regions. Tracking approach is also applied for frame difference, and the generated output motion mask is used to support background model subtraction, especially for slow-moving object cases which cause failure in our background model. The main novelty of the paper is that it proposes a reasonable solution for false detection issue due to homography error without adding a heavy computational cost. We measure each module’s performance to demonstrate the impact of each module on the proposed method clearly. Experimental results are examined on two publicly available aerial image datasets, PESMOD and VIVID. The proposed method runs in real time and outperforms existing background modeling-based methods. It is seen that the proposed method achieves a significant reduction in false positives and has stable performance on different kinds of images.

Langue: Anglais