Hand Gesture Recognition Using Convolutional Neural Networks

• Published in: 2024 Asian Conference on Intelligent Technologies (ACOIT) pp. 1 - 5
• Main Authors: Madaan, Vijay, Sharma, Neha
• Format: Conference Proceeding
• Language: English
• Published: IEEE 06.09.2024
• Subjects: Accuracy; Convolutional Neural Network (CNN); Convolutional neural networks; Data Augmentation; Deep learning; Dropout; Gesture Classification; Gesture recognition; Gray-scale; Grayscale Images; Hand Gesture Recognition; Hands; Hyperparameters; Image Classification; Lighting; Max Pooling; Pattern Recognition; ReLU Activation; Resilience; Training; Transfer learning
• ISBN: 9798350374933
• DOI: 10.1109/ACOIT62457.2024.10939661

This work presents a Kaggle Leap GestRecogn Dataset convolutional neural network (CNN) model for hand gesture recognition. Three Conv2D layers in the model with matching 32, 32, and 64 filters are defined by ReLU activation and max pooling respectively. Using dense layers totaling 256 units and a dropout rate of 0.3, the model produced an amazing \mathbf{99. 97 \%} test accuracy and a test loss of 0.0019. Key hyperparameters are a fifty by fifty picture size, a thirty-two batch size, and seven training cycles. The challenge is precisely matching hand gestures from grayscale images. Sometimes traditional methods cannot allow notable variation in hand placements and lighting conditions. This work significantly improves gesture detection performance using a deep learningbased CNN model with suitable hyperparameters and architectural choices, therefore obtaining remarkable accuracy and resilience. This method considerably increases gesture identification accuracy, therefore enabling deep learning methods to address difficult pattern recognition problems in hand gesture datasets.