A Gesture Recognition Framework Based on Multi-frame Super-resolution Image Sequence

This paper presents a gesture recognition framework based on multi-frame super-resolution image sequence, aiming at solving the low recognition efficiency problem caused by the motion blur during fast gesture transformation and the complex variable visual backgrounds in military operation scenarios. The framework includes three modules: the gesture detection and alignment, the features extraction and fusion, as well as the gesture reconstruction and recognition. Through these three modules, the gesture within each image of the image sequence can be quickly located and aligned, and then the feature information of multiple images can be fused in pixel-level. Finally, the fused features are used to build a gesture reconstruction module to perform high-resolution gesture reconstruction and recognition. On top of that, we build a new video hand dataset named PHV1000, which covers more than a dozen military operation gestures and fills the gap in this field, and we demonstrate the superior performance of our method in this dataset.