Architecture Design and VLSI Implementation of 3D Hand Gesture Recognition System

Area of Science:

• Computer Vision
• Hardware Design
• Human-Computer Interaction

Background:

• Gesture control systems offer intuitive interfaces, replacing traditional devices like keyboards and mice.
• Existing vision-based hand gesture recognition systems face challenges in noisy environments and accurate object extraction.
• Pre-processing methods like skin or motion filters are often insufficient for robust hand detection.

Purpose of the Study:

• To propose a VLSI design for a dual-camera system capable of constructing depth maps and recognizing hand gestures.
• To develop an adaptive depth filter for effective foreground object segmentation in complex environments.
• To enable both static and dynamic hand gesture recognition using depth and coordinate information.

Main Methods:

• A stereo matching algorithm was employed to generate depth maps from dual-camera input.
• An adaptive depth filter was utilized to distinguish foreground objects (hands) from the background.
• Dynamic gesture recognition was achieved by integrating depth and coordinate information.

Main Results:

• The proposed system successfully performs both static and dynamic hand gesture recognition.
• The Application-Specific Integrated Circuit (ASIC) design was implemented using TSMC 90 nm technology.
• The system demonstrated an average accuracy of 83.98% for gesture recognition with low power consumption (27.8 mW).

Conclusions:

• The developed VLSI system offers a robust solution for hand gesture recognition, overcoming limitations of traditional vision-based methods.
• The adaptive depth filtering and dual-camera stereo matching approach enhance accuracy in diverse conditions.
• This hardware-efficient design paves the way for more intuitive and convenient human-computer interaction through advanced gesture control.