Saliency-Driven Hand Gesture Recognition Incorporating Histogram of Oriented Gradients (HOG) and Deep Learning.

Area of Science:

• Computer Vision
• Human-Computer Interaction

Background:

• Hand gesture recognition is crucial for human-machine communication.
• Complex backgrounds challenge existing hand shape recognition methods, particularly those relying solely on skin color.
• Accurate hand region detection is vital for reliable gesture interpretation.

Purpose of the Study:

• To develop a novel and efficient hand gesture recognition model.
• To improve accuracy in complex image scenes and varied conditions.
• To evaluate the model's robustness against background complexity and image noise.

Main Methods:

• Utilized computer vision techniques incorporating saliency maps, histogram of oriented gradients (HOG), Canny edge detection, and skin color.
• Developed an efficient hand posture detection model integrating these features.
• Introduced noise to approximately 60% of the datasets for robustness testing.

Main Results:

• Achieved over 99% accuracy on the NUS Hand Posture Dataset II.
• Reached more than 97% accuracy on a challenging hand gesture dataset with complex backgrounds.
• Demonstrated robust performance with over 98% and nearly 97% accuracy on NUS and hand gesture datasets, respectively, after adding noise.

Conclusions:

• The proposed model significantly enhances hand gesture recognition accuracy in complex environments.
• The integration of saliency maps with HOG provides stable performance across diverse image conditions.
• This approach offers a reliable solution for advanced human-computer interaction systems.