Real-Time Detection of Cyber-Physical Attacks on Smart-IP-Camera Using Network and Telemetry Data

Cyber-physical attacks on edge devices involve exploiting vulnerabilities in both the cyber (software, network) and physical (hardware) components of these devices. The consequences of these attacks include compromising user privacy and intruding into devices, leading to physical damage or disruption of services. Smart cameras are such devices as a part of the broader Internet of Things (IoT) ecosystem and are susceptible to various cyber-physical security threats. To mitigate these cyber-physical threats, manufacturers and users implement security best practices, including regular software updates, strong authentication mechanisms, encryption, and physical security measures. However, traditional methods are insufficient to detect anomalies in the face of evolving cyber-physical threats in real-time as they often rely on predefined rules. Machine learning (ML) based approaches learn from the data and adapt to changing attack patterns, enhancing the security of smart cameras against cyber-physical attacks. In this work, an ML-based adaptive security framework is proposed for smart cameras to dynamically adjust security measures in real-time processing for quick threat detection and immediate responses in case of security breaches or anomalies. A novel model, termed IALSTM (Inference Aware LSTM), is proposed to effectively implement edge-based smart security measures on resource-constrained edge computing devices. Additionally, classical ML algorithms are employed to analyze metrics related to environmental conditions.