Real-time gesture recognition by learning and selective control of visual interest points

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Machine Learning

Background:

• Gesture recognition systems often struggle with real-time performance and accurately focusing on relevant visual information.
• Challenges include handling variations in scale, rotation, clothing, and individual motion characteristics.

Purpose of the Study:

• To develop a comprehensive framework for real-time recognition of unspecified gestures from arbitrary individuals.
• To address the critical issues of selective attention and processing frame rate in gesture recognition systems.

Main Methods:

• Proposed the Quadruple Visual Interest Point Strategy for robust feature extraction from dynamic regions of interest, assigning probability density functions to visual interest points.
• Developed a selective control method for self-load monitoring and control within the recognition system to manage processing frame rate.
• Implemented a gesture video system to demonstrate real-time performance and interactive capabilities.

Main Results:

• The Quadruple Visual Interest Point Strategy enables recognition without assumptions on scale or rotation of visual features.
• The selective control method ensures the system's self-load monitoring and controlling functionality for efficient processing.
• Evaluation experiments confirmed robust recognition across diverse factors like clothing, gesture type, motion trajectories, and individual differences.

Conclusions:

• The presented framework effectively addresses key challenges in real-time gesture recognition, improving accuracy and speed.
• The system demonstrates practical utility and real-time performance through a developed gesture video system.
• This approach offers a significant advancement for natural and intuitive human-computer interaction through gesture control.