Detection of Dynamic Frame Alteration in IoT Video Streams

The rapid expansion of IoT video systems in security-critical environments has increased exposure to sophisticated video manipulation attacks. Traditional detection methods including machine learning approaches such as SVM, kNN, and LOF show strong performance against replay, frame injection, and stream hijacking attacks, but fail to detect Dynamic Frame Alteration (DFA), a subtle and adaptive manipulation technique that introduces minimal visual disturbance and leaves limited statistical traces in pixel space. To address this limitation, this work presents a real-time DFA detection framework that relies on system-level behavioral metrics rather than video content analysis. The proposed method monitors frame rate, CPU usage, memory load, and process activity using sliding time windows, rolling averages, and empirically validated thresholds, enabling robust detection. Experimental results show that DFA manipulation produces consistent deviations in these system metrics,particularly frame drops and CPU/memory spikes which are effectively captured by the proposed approach. Across two realistic scenarios the detector achieves up to 93% detection accuracy, significantly outperforming traditional machine learning models whose detection rates on DFA average below 25%.These findings establish system-metric analysis as a practical and explainable alternative to machine learning methods to detect advanced video manipulation in IoT environments.