Navigating Virtual Environments Using Leg Poses and Smartphone Sensors.

Area of Science:

• Human-Computer Interaction
• Virtual Reality
• Machine Learning

Background:

• Virtual environment navigation presents significant challenges due to complex operating conditions.
• Effective translation of physical actions to virtual movements requires sensor fusion and machine learning.
• Identifying optimal sensor data and machine learning algorithms is crucial for realistic virtual interactions.

Purpose of the Study:

• To develop an innovative approach for virtual environment navigation using readily available hardware.
• To synchronize physical leg movements with virtual avatar actions.
• To select significant features and appropriate machine learning algorithms for movement identification.

Main Methods:

• Utilized smartphone sensors (gyroscope, accelerometer, compass) attached to the lower leg.
• Implemented data pre-processing using the box plot outliers approach.
• Employed machine learning techniques, including Artificial Neural Networks, for movement identification.

Main Results:

• Artificial Neural Networks demonstrated the highest movement identification accuracy.
• Achieved 84.2% accuracy on the training dataset and 84.1% on the testing dataset.
• Successfully translated identified leg movements into virtual avatar navigation.

Conclusions:

• The proposed approach enables intuitive virtual environment navigation via simple leg movements.
• Smartphone-based sensor data combined with machine learning offers a viable solution for virtual locomotion.
• Artificial Neural Networks provide a robust algorithm for accurate movement identification in this context.