HandFormer: A Dynamic Hand Gesture Recognition Method Based on Attention Mechanism

The application of dynamic gestures is extensive in the field automated intelligent manufacturing. Due to temporal and spatial complexity gesture data, traditional machine learning algorithms struggle extract accurate features. Existing recognition have complex network designs, high parameter counts, inadequate feature extraction. In order solve problems low accuracy computational current recognition, a model based on MetaFormer architecture an attention mechanism was designed. proposed fuses CNN (convolutional neural network) Transformer by embedding convolution into model. Specifically, token mixer block replaced Spatial Attention Convolution Block Temporal obtain Former Block. Firstly, each frame input image quickly down-sampled PoolFormer then learn information. Then, maps learned from are concatenated along channel dimension information action. Finally, overall classified category gestures. achieves average 96.72% 92.16% two publicly available datasets, Jester NVGesture, respectively.