MediaPipe Landmark-Based Inter-Frame Gesture Stability Determination Algorithm

Gesture recognition technology plays a crucial role in real-time interaction scenarios such as non-contact control in intelligent cockpits and AR/VR applications. However, traditional static gesture recognition methods suffer from high computational resource consumption and misclassification rates in dynamic scenarios. This study proposes a MediaPipe Landmark-based inter-frame gesture stability determination algorithm, which dynamically schedules the inference timing of static gesture recognition models by monitoring real-time inter-frame gesture variation metrics and designing a state transition mechanism. For our experiments, we employ a YOLOv8 static gesture recognition model trained on the HaGRID dataset. To evaluate performance in dynamic scenarios, we construct a dynamic cross-domain test set comprising 133 video samples across 6 categories via semantic consistency mapping on the NVGesture dynamic gesture dataset. Results demonstrate that the algorithm reduces the dynamic misclassification rate by 45.20%, with a per-frame processing latency of only 26.67 ms in a CPU environment, meeting the 30 FPS real-time requirement. This method requires no additional training costs and can theoretically be directly adapted to most existing static gesture recognition models, providing an efficient solution for real-time gesture interaction in dynamic scenarios.