DEVELOPMENT OF A VIRTUAL ENVIRONMENT BASED ON EEQ SIGNALS
TL;DR: A bio-cybernetic adaptation system for interactive virtual environments, driven by real-time human emotion recognition using Electroencephalography (EEG) signals, which is fully compatible with future direct integration into Virtual Reality (VR) technologies.
We presents the development of a bio-cybernetic adaptation system for interactive virtual environments, driven by real-time human emotion recognition using Electroencephalography (EEG) signals. The research utilizes the internationally recognized DREAMER dataset, which contains multimodal biometric recordings from 23 participants. An efficient machine learning pipeline was constructed for EEG signal processing. After artifact removal and normalization, the raw signals were segmented into 1-second sliding windows (epochs) to ensure real-time responsiveness, yielding 85,744 unique training samples. From each epoch, 878 comprehensive features were extracted across both time-domain (Mean, Std, Hjorth parameters, Skewness, Kurtosis) and frequency-domain statistics (Power Spectral Density via Welch's method, spectral entropy, edge frequencies, and inter-channel correlations). A Random Forest classifier was deployed for a 3-class emotional state classification (Negative, Neutral, Positive) along the Valence axis, achieving an overall accuracy of 70.39% on the test dataset. The trained model was serialized into rf_model.pkl and integrated into a Python backend. This backend streams predictions via UDP protocol to a Unity-based 3D virtual environment, dynamically modifying the visual atmosphere and acoustic crossfading in synchronization with the user's internal emotional state. The developed software architecture is fully compatible with future direct integration into Virtual Reality (VR) technologies.