Hough Forest With Optimized Leaves for Global Hand Pose Estimation With Arbitrary Postures.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Machine Learning

Background:

• Robust global hand pose estimation is crucial for intuitive human-computer interaction.
• Existing methods often struggle with accuracy and robustness, particularly with single depth cameras.
• Full degree-of-freedom hand pose estimation remains a challenging research area.

Purpose of the Study:

• To develop a novel algorithm for optimizing Hough forest leaf weights for improved global hand pose estimation.
• To enhance the accuracy and robustness of hand pose estimation using a single depth camera.
• To suppress ambiguous votes in prediction fusion for more reliable results.

Main Methods:

• A novel algorithm to optimize leaf weights in a Hough forest was developed.
• Leaf node weights were learned to minimize average pose prediction error.
• The approach was tested on both synthetic and real-world depth datasets.

Main Results:

• The proposed method significantly improved hand pose estimation accuracy compared to traditional Hough forests.
• Optimized leaf weights led to superior performance against state-of-the-art convolutional neural network-based methods.
• The algorithm demonstrated enhanced robustness on real-world depth videos, outperforming other hand tracking systems.

Conclusions:

• The optimized Hough forest algorithm provides a principled and effective approach to global hand pose estimation.
• The method offers improved accuracy and robustness, making it suitable for real-world applications.
• The technique enables the development of intuitive virtual/augmented reality applications for bare-hand interaction.