QStab: A lightweight video stabilization algorithm robust to high-frequency perturbations

In this study, QStab, a novel online, real-time video stabilization method is introduced, specifically designed to be robust against high-frequency perturbations commonly encountered near internal combustion engines or other sources of rapid mechanical vibrations. QStab employs a lightweight queue-based mechanism. This method, termed “frame mirroring”, identifies and displays the frame from the buffer exhibiting the minimum mean-squared error relative to an adaptively updated reference frame, thereby suppressing jitter while maintaining visual continuity. Experimental evaluations on the challenging benchmark datasets (SBMNet and NUS) demonstrate that QStab consistently achieves a perfect cropping ratio of 1.0 across all test categories, thereby pre-serving full-frame content, a key advantage over existing methods. Additionally, it delivers a strong inter-frame transformation fidelity (ITF) score and competitive distortion metrics, confirming both temporal smoothness and geometric consistency. Although its stability score is slightly lower than state-of-the-art deep learning methods, the overall performance reflects a well-balanced trade-off between visual quality and computational simplicity. Given its low complexity and real-time execution, QStab is particularly suitable for embedded systems with stationary cameras and real-world surveillance applications in vibration-intensive environments. The source code of our proposed method as well as the link side-by-side video comparisons demonstrating the algorithm’s per-formance can be found in our GitHub page.