An Efficient Intrusion Detection Approach for Visual Sensor Networks Based on Traffic Pattern Learning

Visual sensor networks (VSNs) are highly vulnerable to attacks due to their open deployment in possibly unattended environments. To improve the network security of VSNs, an intrusion detection system (IDS) is an effective countermeasure. However, as visual sensors can produce big and dynamic video data, it is a tough task to rapidly and effectively detect attacks in VSNs. Moreover, attack samples in VSNs are generally too rare for IDSs to fully understand the behaviors of attacks. Facing these difficulties, in this paper, we propose an efficient intrusion detection approach for VSNs, which is based on traffic pattern learning. In the proposed approach, a traffic model is developed to describe the dynamic characteristics of network traffic in VSNs. Based on this model, the optimal feature set for traffic pattern learning can be extracted. Then a hierarchical self-organizing map (HSOM) is employed to learn traffic patterns and detect intrusions. Furthermore, an active learning strategy is devised to accelerate the training process of the HSOM and better learn the patterns of attacks. Experimental results show that the proposed approach has high detection accuracy and good real-time performance.