D-Resnet: A Static Gesture Segmentation Recognition Network Based on Feature Fusion

Nowadays, gesture recognition based on vision plays an increasingly important role in fields such as augmented reality. However, most of the existing researches focus on the ideal lighting conditions, while gesture recognition in complex lighting conditions is not highly concerned, which also leads to the low recognition rate in this scene. To address the above problems, this paper proposes a static gesture segmentation recognition network with feature fusion: first, the input image is pre-processed with noise, then the feature map of the image is extracted by using DeepLabV3, then the pre-processed image is masked by the obtained feature maps, and the masked result is fused with the pre-processed image as the input of the network, and then Resnet18 is used for gesture recognition. The experimental results in both datasets show that the proposed method in this paper has higher recognition accuracy under different lighting conditions compared to Resnet and its variants.