Hybrid Efficientnet - Data-Efficient Image Transformer Model for ASL Hand Gesture Recognition
Ishan Chaskar, P Santhiya, Stewart Kirubakaran S, V. Ebenezer, Manoah Noble · 2025
A valuable information process is important because the complexity of the description depends on small features such as finger placement, hand orientation, and textures. This research addresses complex tasks of cognitive processing that are important in fields such as videography, remote surgery, and sign interpretation. The work presents the most innovative methods for cognitive performance using a hybrid model consisting of transfer learning and vision transformers that focus on segmentation, output concatenation, inference, and interconnection. By analyzing different data containing 26 gestures captured through Mediapipe in an unstructured environment to map the essential key points and filter out the unnecessary background information, the proposed system has leveraged the use of the pre-trained transfer learning model EfficientNet along with Data-efficient Image Transformer to effectively process hand gestures, ensuring robust recognition. The combination of these two makes the model suitable for real-world applications requiring precise hand gesture detection, such as interactive systems and sign language interpreters. The proposed model achieves an impressive accuracy of 97 %, demonstrating its effectiveness in recognizing hand gestures with high precision.