Abstract
In the context of video surveillance applications in outdoor scenes, the moving object classification still remains an active area of research, due to the complexity and the diversity of the real-world constraints. In this context, the class imbalance object distribution is an important factor that can hinder the classification performance and particularly regarding the minority classes. In this paper, our main contribution is to enhance the classification of the moving objects when learning from imbalanced data. Thus, we propose an adequate learning framework for moving object classification fitting imbalanced scenarios. Three series of experiments which were led on a challenging dataset have proved that the proposed algorithm improved efficiently the classification of moving object in the presence of asymmetric class distribution. The reported enhancement regarding the minority class reaches 116% in terms of F-score when compared with standard learning algorithms.