Eye-Nertia

[AshLu]

Eye-Nertia

Yazan N. Billeh

Victor Krug

Brenna Smith

Introduction

Spinal cord injuries often result in paralysis of all four limbs and the trunk of the body is a disorder known as tetraplegia (also

known as quadriplegia). In the United States over a quarter of a million people suffer from spinal cord injuries, approximately 20%

of which have complete tetraplegia [1]. Investigations have shown, however, that individuals with such disorders and disabilities

still maintain an intact retina and normal control of their oculomotor system [2]. This has motivated us to develop an eye-controlled

method to control a wheelchair via the use of electrooculography (EOG). Since time and budget constraints did allow us to integrate

our control system with an electronic wheelchair, we developed a proof-of-concept prototype to show how EOG signals can control

a motorized vehicle with a remote control. The system hardware was composed of standard, off-the-shelf, electronic components

while the software and control was implemented with LabVIEW software. The system developed allows theoretical movement at

different speeds (the remote control car used works on an ON-OFF basis and so the car itself travelled at a single speed), 360

o

rotations, and is capable of an emergency stop. In order to meet our design criteria, our device was tested on simple driving tasks,

such as backwards parking, and required that the driver successfully complete these tasks in a time not more than 1.5 times longer

than the time recorded for completing them by remote control.

Experimental Setup

The signal acquisition and processing portion of the system utilized four electrodes placed on the face of the patient. These

disposable electrodes were placed on the left and right temple adjacent to the eyes and directly above and below the left eye (Figure

1). The difference between the horizontal electrodes and the difference between the vertical electrodes was processed by several

circuits. One of the first circuits investigated involved passing the EOG signals through a differential amplifier followed by a highpass

filter and finally a low-pass filter. Due to baseline drifts, a long-time constant high-pass filter was tested [3]. The circuits gave

excellent results in one direction of eye movements. However, since only one direction was suitably detected, four inputs would

have been needed for the DAQ board (since a reference at the forehead would be required). To avoid the excess wires and make the

system more comfortable for the user, we simplified our circuit by removing the high-pass filter. Although this resulted in baseline

drifts, it was decided that this issue will be resolved with the LabVIEW VI. The final circuit used provided a gain of 1000 and the

low-pass filter had a cut-off of 100Hz (parameters chosen in accordance to [4]). The two voltages recorded were then acquired by

LabVIEW via the DAQ board.

The software interface of the system was designed to be intuitive and easy to calibrate. The five thresholds could be easily

determined after a short observation period. These thresholds were set to a value slightly lower or higher than the voltage generated

by a blink or the movement of the eye left, right, up, or down. If one of the directional thresholds were to be exceeded, the

LabVIEW program would output a voltage the controller circuit in order to increase the velocity or rotation of the wheels.

Exceeding the blink threshold would cause the emergency stop function to execute, which would turn off all outputs until

reactivation of the LabVIEW program (a “heavy” blink was required for the emergency stop). The four outputs of the LabVIEW

program were connected to the controller circuit via two DAQ boards. The controller circuit consisted of the remote control with

which wires were soldered onto its circuit board and four MOSFET transistors (N-MOS) were connected to these wires on a

breadboard (Figure 2). Each wire was connected to the ground of the controller and one of the four directional switches on the

remote control. The output voltage of the software interface was connected to the gate of the N-MOS to pass a voltage value that

surpasses the transistor’s threshold voltage to activate the transistors. Therefore, when the software outputs a voltage, the

directional circuit is completed and the remote controller outputs a signal as if that directional button was pressed. The remote

control car (our budget wheelchair) then moves in the manner that was designed by the toy company.

Results

The system developed is functional and can be easily controlled by the user’s eye-movements once the thresholds are set (a process

that requires approximately 1 - 3 minutes). An example of the LabVIEW VI front panel that confirmed our recording set-up is

functional is shown in Figure3.A and screen shot of the LabVIEW VI block diagram is shown in Figure 3.B. The user simply moves

their eyes and the speed and rotation scales are observed to see the responsiveness of the system. In addition, the output from the

DAQ board was confirmed to be functional by checking which N-MOS transistors are on when the DAQ was outputting a voltage

in response to the signal analysis of the LabVIEW VI. This was checked in addition to the actual voltage output given by the DAQ

board to confirm that the output voltage was dependent on the speed and rotation selected by the user. Although there was

electromagnetic interference (EMI) when the complete system of both recording EOG signals and controlling the car was set-up,

this issue was resolved by a “homemade” Faraday cage and the system was functional. The car was difficult to control at first;

however with approximately 4±1 hours (not necessarily consecutive hours), the user was able to attain adequate control of the car to

perform certain tasks. This learning time will be greatly reduced with a slower car (wheelchair) as the user will not overshoot or

undershoot when pursuing a certain track.

Our original goal was to have the automobile complete a figure eight track and a parallel park in a time frame exceeding no more

than 1.25 the time required to complete these tasks with the remote control. This, however, had to be modified solely due to the

capabilities of the remote control car. The car had a variable (but large) turning radius and was too fast. By using weaker batteries to

slow down the car, it was no longer possible to turn the car. Adding weights to the car was also attempted but this was ineffective as

it eliminated the car’s ability to turn. In addition, to achieve different speeds, we modified the LabVIEW code to have the DAQ

board give an output voltage to the transistors in a pulsing fashion and varying the duty cycle of the pulses in accordance to the

selected speed. However, this greatly lengthened the refresh rate of the code (time to complete a loop) and hence there was a large

time-delay between the users’ selection and the response of the system. Thus, we were not able to effectively slow down the car

while maintaining the functionality of the car. Hence, it was extremely difficult to perform our initially proposed testing tasks using

the remote control itself. As a result, the acceptance criteria had to be modified to a goal reasonable to the car’s performance. The

car was able to perform a reverse park (Figure 4) and complete a circular track without any issues. The time required to complete a

reverse park took on average approximately 1.5 times the required time to perform this task with the remote control. The circular

track, however, may be argued to be easier with the EOG-system as once the user sets the rotation and speed, the system will

complete the track indefinitely, while with the remote control, the user is required to manually keep the buttons pressed.

Overall, the system developed does meet the acceptance criteria set forth by allowing the user to have complete control of the

remote control car using eye-movements. The aim of developing a proof of concept system to show that our developed system is

suitable to be integrated to a real wheelchair has been established. In addition the goals set (in accordance with the car’s abilities)

were accomplished. In general, our device is also quite reliable. A questionnaire made (Figure 5) indicated that users felt they had

control over the car and felt safe using the system (due to the emergency stop function). As with any engineering system, certain

limitations and reliability issues are present with our prototype. For instance, although the LabVIEW code accounts for baseline

modifications, the signal amplitudes may change with time and recalibration may be needed if the electrode gels begin to dry out.

This rarely occurred but is certainly an issue if the device is to be utilized by patients for day long periods. In addition, the

emergency stop function would completely halt the system and hence the user would need a method to reactivate this, for instance,

vocally. The system would also sporadically appeared to be picking up noise due to EMI, but this would not be an issue in a

wheelchair as no long distance transmission is required as the user will move with the wheelchair.

Discussion and Conclusion

We were able to successfully create a proof of concept model for an EOG-controlled wheelchair and to satisfy our modified design

criteria with this model. Users were able to demonstrate full linear control of their speed and rotation as demonstrated by their

manipulations of the gages on the front panel of our LabView VI, as well as control of the car. We also implemented a successful

emergency stop system. Additionally, users able to perform a backwards park with our system, and completed this in a time ~1.5

longer than the time recorded by performing this with the remote control. Our proof of concept model may be significantly

improved by the use of a model car that travels at a slower speed. This would enable the user to actively manipulate the car so

further testing and validation of our model could be conducted. Our project had severe limitations due the speed of the car, which

disabled us from completing the desired testing due to the users’ lack of control of the system. The project was also limited in that

the users had to monitor the screen to track their speed, and, thus, were not able to observe where they were driving the car, making

even the simplest driving tasks exceedingly difficult.

Application

The next step in the development of this system is to integrate it into an electric wheelchair. We have demonstrated that the motion

of subject’s eyes can generate signals that may be used to guide their motions. Integration into a wheelchair would allow the user to

see where they were going, thus eliminating this major drawback in our proof of concept model. The speed of the wheelchair would

also be a practical and safe speed that would allow the subject the necessary manipulation of the wheelchair to ensure efficacy and

safety, and thus eliminating the other major drawback in our proof of concept model. Integrating wireless electrodes would

minimize the impedance of the system, specifically the electrodes, on the user’s daily activities. Another addition to the system

would be to have the user control the speed linearly and not simply at discrete speeds and turning angle. The system would

ultimately be useful for the disabled tetraplegic population highlighted in the introduction, serving as an aid in restoring

independence and mobility to this population. This will improve the quality of life of these individuals as they would require less

assistance which is of great interest to the overstretched health care system.

To implement this system and allow it to become commercialized, it will be necessary build a prototype with a real wheelchair.

Following this, institutional review board (IRB) approvals would be sought to test the device. If successful, venture capitalist would

be contacted to attain funding to begin a start-up company to manufacture these wheelchairs (or contacting existing companies to

sell them the idea is also possible). In addition, patents would be applied for whenever possible that are unique for our design. For

instance, current systems use 5 electrodes and 4 inputs [4] but by modifying our circuit and electrode placement to only require 2

inputs, it is conceivable that this idea may be patentable which will increase the possibilities of successful commercialization

(further research is required however to ensure this is patentable).

References

[1] “Spinal Cord Injury Facts & Figures at a Glance.”

National Spinal Cord Injury Statistical Center. 2008. University of Alabama.

3 April 2010

[2] Ferman, L., Collewijn, H., Jansen, T. C., and van den Berg, A. V. (1987) Human gaze stability in the horizontal, vertical and

torsional direction during voluntary head movements, evaluated with a three-dimensional scleral induction coil technique,

Vision

Research

27: 811–828.

[3] Lawrence Y. Deng , Chun-Liang Hsu, Tzu-Ching Lin, Jui-Sen Tuan, and Shih-Ming Chang (2010) EOG-based Human–

Computer Interface system development.

Expert Systems with Applications 37: 3337–3343

[4] Barea, R., Boquete, L., Mazo, M., and López, E. (2002) Guidance of a wheelchair using electrooculography.

Journal of

Intelligent and Robotic Systems

34: 279–299

Attachment: Original Report