Hand Pose Recognition Using Parallel Multi Stream CNN.

Area of Science:

• Computer Vision
• Machine Learning
• Human-Computer Interaction

Background:

• Computer applications increasingly utilize hand gestures for input, moving beyond traditional interfaces like keyboards and touchscreens.
• Existing methods using RGB data with Support Vector Machines and Neural Networks show limitations in performance for hand-pose recognition.
• Depth data offers improved understanding of posture attributes, presenting an opportunity for enhanced hand gesture recognition.

Purpose of the Study:

• To propose and evaluate a novel multiple parallel stream 2D Convolutional Neural Network (CNN) model for accurate hand posture recognition.
• To leverage depth data for improved detection of hand poses in various computer applications.
• To compare the proposed model's performance against existing state-of-the-art methods.

Main Methods:

• A multiple parallel stream 2D CNN model was designed, incorporating multiple steps and layers for hand pose detection from depth data.
• Model hyperparameters were optimized through rigorous experimental analysis.
• The model was trained and tested on three diverse, publicly available benchmark datasets: Kaggle, First Person, and Dexter.

Main Results:

• The proposed model achieved exceptional accuracy rates of 99.99% (Kaggle), 99.48% (First Person), and 98% (Dexter).
• Near-optimal F1 and Area Under the Curve (AUC) scores were obtained, indicating robust performance.
• Comparative analysis demonstrated that the proposed model significantly outperforms previous state-of-the-art methods.

Conclusions:

• The developed multiple parallel stream 2D CNN model effectively recognizes hand postures using depth data with high accuracy.
• This approach represents a significant advancement in hand-pose-based computer applications, offering superior performance.
• The model's effectiveness across multiple datasets validates its potential for real-world applications in gesture recognition.