Real-time depth-based hand detection and tracking

Area of Science:

• Computer Vision
• Robotics
• Human-Computer Interaction

Background:

• Real-time hand detection and tracking are crucial for interactive systems.
• Existing methods often struggle with depth data variations and accuracy.

Purpose of the Study:

• To develop an efficient and accurate real-time hand detection and tracking system using depth data.
• To improve upon existing hand tracking algorithms by incorporating depth-specific features and adaptive algorithms.

Main Methods:

• A boosting-cascade classifier is proposed for hand region detection, utilizing depth-difference features.
• A region growing scheme is employed for tracking hands across successive frames, initiated from a seed point.
• The Depth Adaptive Mean Shift (DAM-Shift) algorithm is introduced to accurately determine the hand's central point, adapting its search based on hand depth.

Main Results:

• The proposed method demonstrates effective real-time hand detection and tracking capabilities.
• Qualitative and quantitative comparisons show competitive or superior performance against existing AdaBoost methods.
• Performance analysis across various scenarios validates the tracking accuracy.

Conclusions:

• The developed hand detection and tracking algorithm offers a robust solution for real-time applications using depth data.
• The integration of depth-difference features and the DAM-Shift algorithm enhances localization and tracking precision.
• The method shows promise for advancing human-computer interaction and robotic applications.