Design and Evaluation of an Alternative Control for a Quad-Rotor Drone Using Hand-Gesture Recognition

Area of Science:

• Robotics and Human-Computer Interaction
• Computer Vision and Machine Learning

Background:

• Hand-gesture recognition (HGR) has been researched for 40 years, with modern advancements in machine perception enabling algorithms like MediaPipe Hands (MPH).
• Existing HGR solutions vary in medium, method, and application, necessitating evaluation for new control contexts.

Purpose of the Study:

• To evaluate the applicability of modern HGR algorithms, specifically MPH, for alternative control.
• To develop and assess an HGR-based system for controlling a quad-rotor drone.

Main Methods:

• Evaluation of MediaPipe Hands (MPH) for hand-gesture identification using a single-camera, skeletal model approach.
• Development of a novel HGR algorithm by selecting a classifier to complement MPH's lightweight nature and address its Z-axis instability.
• Implementation of the HGR system for controlling a quad-rotor drone.

Main Results:

• MPH exhibited Z-axis instability, reducing landmark accuracy from 86.7% to 41.5%.
• The selected classifier compensated for MPH's instability, achieving 96.25% classification accuracy for eight static single-hand gestures.
• The developed HGR system enabled intuitive, computationally inexpensive, and repeatable drone control.

Conclusions:

• Modern HGR algorithms like MPH can be adapted for alternative control applications.
• The developed HGR system offers a viable solution for accessible and efficient drone operation without specialized equipment.