This work focuses on the use of brain signals to interpret the neurophysiological signals as a direct communication pathway to an external device. The main objective is to build a simple development platform based on a commercially available 14-channel electroencephalography (EEG)-based brain-computer interface (BCI) to control a 3D game. The proposed method consists on harvesting, recording and processing of EEG data from an Emotiv Epoc+ headset. Support vector machine (SVM), linear neural network (NN) and decision trees (DT) are used to build machine learning models and are compared based on statistical measures. The models are trained on two control states: forward and backward, in order to control the character in the game. In the first stage of modeling, each user is built using its own unique model based on the data gathered during user training. Then a more general model is developed to allow plug and play interactions with many users, without the need for retraining. The proposed classifiers are able to sufficiently model brain control for a single user with acceptable errors for the motor action training dataset, but fail to generalize to datasets with multiple users. Feature extraction improves learning of the models like neural network, but decreases the number of learning samples which calls for more datasets to generalize results for multiple users. The 3D game used for control is built on Unity 5 game engine. The traditional navigation controls in the game are mapped to the machine learning model predictions, while the internal Gyroscope is used for panning and rotation.
Paper
Full text
EEG-based Control of a 3D Game Using 14-channel Emotiv Epoc+
Semantic Scholar · Computer Science · 2019
Abstract
This work focuses on the use of brain signals to interpret the neurophysiological signals as a direct communication pathway to an external device. The main objective is to build a simple development platform based on a commercially available 14-channel electroencephalography (EEG)-based brain-computer interface (BCI) to control a 3D game. The proposed method consists on harvesting, recording and processing of EEG data from an Emotiv Epoc+ headset. Support vector machine (SVM), linear neural network (NN) and decision trees (DT) are used to build machine learning models and are compared based on statistical measures. The models are trained on two control states: forward and backward, in order to control the character in the game. In the first stage of modeling, each user is built using its own unique model based on the data gathered during user training. Then a more general model is developed to allow plug and play interactions with many users, without the need for retraining. The proposed classifiers are able to sufficiently model brain control for a single user with acceptable errors for the motor action training dataset, but fail to generalize to datasets with multiple users. Feature extraction improves learning of the models like neural network, but decreases the number of learning samples which calls for more datasets to generalize results for multiple users. The 3D game used for control is built on Unity 5 game engine. The traditional navigation controls in the game are mapped to the machine learning model predictions, while the internal Gyroscope is used for panning and rotation.