Real-Time Control of a Video Game Using Eye Movements and Two Temporal EEG Sensors

Table 6

Advantages and disadvantages of eye movements classification based on EEG signal.

Criteria	Advantage	Limitation
Visual angle	A small visual angle between 5° and 10° was used to decrease fatigue issue (a large visual angle of 30° or more is required to detect eye movement in most research using EOG signals. This large visual angle leads almost immediately to eye fatigue, exhausting the user).	It becomes difficult to detect eye movements if the visual angle is less than 5°.
User	Several participants were tested (offline [20], online [21], and in different real-time experiments in this study) on different days to examine the variability and nonstationary nature of EEG signals.	Absence of testing the proposed algorithm on handicapped users.
Sensors position & number	(i) The position of sensors around the ears is more robust to muscles activity noise (body or head movements do not influence so much the classification accuracy).; (ii) Two temporal EEG sensors were used (4 attached sensors on the face are used as minimum requirement to get good classification accuracy in EOG technique).	A low-cost wireless device based on the proposed idea is not yet developed.
Comfort and portability	The most suitable sensors position for daily life applications to record eye movements compared with EOG sensors (the sensors can be attached to the end of the glasses arms (temples), headset, and headband).	Less comfort [21].
Real-time classification	(i) Single trial was used for real-time classification.; (ii) No training or calibration phase was added before real-time classification (fixed and common thresholds for all subjects were used).; (iii) No fixed time interval for eye movements (the user is free to move his/her eyes and send commands at any moment).; (iv) Six classes were distinguished using a linear clarifier.; (v) Eye movements were detected and classified in open- and closed-eyes cases.; (vi) The proposed algorithm was tested in several real-time scenarios.	Using average or loop to make a decision or machine learning methods can improve the classification accuracy but decrease the response time [9–13, 29].
Real-time control	(i) Asynchronous control (the user can send commands even with closed eyes using noninvasive technique).; (ii) The classification results were used for full control of continuous character’s movement in 2D video game.; (iii) The bit rate for controlling the video game was 30 bits/min.	For each application, we need to develop an interface between classification results and the controlled device.
Classification accuracy	Classification accuracy with chance level of 16.67% was greater than 70%, the suggested minimum for reliable BCI control with chance level of 50% [34].	As same as EOG technique [20].