Sei sulla pagina 1di 4

Advanced Science and Technology Letters

Vol.96 (IRTT 2015), pp.72-75

Method for Measuring 3D Cognitive Load by

Analyzing Dimension Reduced EEG Features

Daejune Ko1, Min Woo Park1, and *Eui Chul Lee1

Department of Computer Science, Sangmyung University, Seoul, Republic of Korea
*Corresponding author:

Abstract. In this paper, a method for measuring 3D cognitive load is proposed.

For that, 8-channel EEG data when watching 2D and 3D displays are 2-
dimensionally reduced using principal component analysis then the reduced
features are visualized and compared. Consequently, we confirmed that the
difference between two feature distributions of watching 2D and 3D at early
stage was significant then the difference was reduced through watching time
flow. The result can be analyzed that the cognitive load caused by 3D binocular
visual combination at early stage and the adaptation of 3D watching time.
Keywords: 3D visual cognitive load, dimension reduction, visualization

1 Introduction

Recently, 3D display fields are rapidly grown. For example, numerous movies and
animations are released and supplied through theater and IPTV. When people are
watching the 3D display, they could feel cyber-sickness which cannot be felt in
watching 2D display. For example, people could be often sensed visual fatigue,
disgust or dizziness during watching the 3D movies or animation. To quantitatively
measure such cyber-sickness, researchers have been studied comparative physical
reactions of the people when watching 2D and 3D display [1][2].
In works, several researches have been performed by using image or signal based
methods. Previous works focused on comparing difference of the amount of fatigues
caused by watching 2D and 3D display. However, these researches were performed
under the assumption that 3D caused more fatigue than 2D. Especially, EEG signal
based research should be focused on the visual cognitive process rather than the
Therefore, EEG temporal tendencies of watching 2D and 3D displays are
visualized and analyzed in this paper. 8-channel EEG signals are 2-dimensionally
visualized by dimension reduction using Principal Component Analysis (PCA). By
comparatively analyzing the tendencies, the visual cognitive processes for watching
2D and 3D can be estimated.

ISSN: 2287-1233 ASTL

Copyright 2015 SERSC
Advanced Science and Technology Letters
Vol.96 (IRTT 2015)

2 Proposed Methods

In this experiment 30 subjects were participated. They had enough sleep and did not
consume caffeine, alcohol and cigarettes in the 12 hours prior to the experiment. In
this experiment, a 30 minutes clip of a 3D enabled movie was used. For this
experiment, the 30 subjects were randomly assigned into two groups that watched the
2D or 3D display respectively. When watching the display, subjects wore a device
which could measure EEG signals and sat a comfortable chair at a 50 cm distance
from the display.
As shown in Fig. 1, EEG signals were obtained from 8-channel (F3, F4, C3, C4, P3,
P4, O1, O2) which are located as four positions per the left and right hemispheres,
respectively. The positions were consisted of frontal lobe, central position, parietal
lobe, and occipital lobe.

Fig. 1. EEG electrodes attaching positions by International 10-20 System [3] and actually
attached positions (red circles) in this research.
Obtained EEG temporal signals were transformed into frequency domain by
performing Short Time Fourier Transform (STFT) at every 1 second region. After that,
the total amount of magnitude at beta bandwidth (14~30Hz) was calculated as feature
which generally means human state of tension or concentration. Consequently, 1,800
(sec) 8 (channels) features of each subject can be obtained.

Fig. 2. Visualization examples of dimension reduced EEG feature for watching 2D and 3D
displays, respectively.

Copyright 2015 SERSC 73

Advanced Science and Technology Letters
Vol.96 (IRTT 2015)

Our method has seven procedures. If 2D signals and 3D signals are visualized on
different space, to compare between two visualized signals is not meaningful.
Therefore to compare between two visualized signals, EEG signals were visualized on
same space when watching 2D and 3D display. So signal sets of two groups were
combined as 3600sec 8-channel. To obtain EVi and evi (in here, i is the number of
EEG channel (= 8)), channels of combined signals were analyzed and 2-
dimensionally reduced by using principal component analysis (PCA) [4][5].
Dimension reduced features by EV7 and EV8 (which have top 2 eigenvalues) can be
represented onto 2D plane as Fig. 2.

3 Results and Discussion

The obtained eigenvalues (evn) and their corresponding eigenvectors (EVn) which
were calculated by performing PCA of 8-channel features of frequency bands. In here,
all features of watching both 2D and 3D were used in order to obtain non-biased
eigenvectors against 2D or 3D features. Among the obtained 8 eigenvectors, only two
eigenvectors should be selected as axes for dimension reduction. Because the 8
eigenvalues and their corresponding eigenvectors were obtained from the 88
covariance matrix, the calculated eigenvalue mean variance of the dimension reduced
features onto the corresponding eigenvector.
After orthogonal projection of features onto an eigenvector, the greater variance
means features well representation because the distance between two dimension
reduced features is greater. Therefore, two eigenvectors having the greatest
eigenvalues such as EV7 and EV8 were selected for visualization of EEG features.

Fig. 3. Visualized feature distributions about watching 2D and 3D displays at every 10mins.
Left column: 0~10mins. Middle column: 10~20mins, Right column: 20~30mins.

EEG signals were visualized on every 10mins for analyzing detail. As shown in
Fig. 3 (b) and (c), visualized signals were not represented difference. Also generally,
fatigue is defined by the reaction which is felt by long-term experiments in body. As

74 Copyright 2015 SERSC

Advanced Science and Technology Letters
Vol.96 (IRTT 2015)

shown in Fig. 3 (a) to (c), visualized 2D signals and 3D signals are getting more
similar as time goes on. As a result, the factors that caused fatigue were almost similar
when watching both 2D and 3D display. However, the difference was represented by
two groups as shown in Fig. 3 (a). Visualized signals were concentrated in watching
3D display more than watching 2D display. In the other words, EEG signals by
stimulation of 3D display than 2D display are focused on a specific position at the
beginning of watching. As a result, a special stimulus was found by watching the 3D

4 Conclusions

In this paper, two dimensional visualization of cognitive feature distribution were

performed by using the acquired EEG data from 8-electrodes when watching 2D and
3D displays. In the experiments, 8-channel EEG data was visualized and analyzed by
dimension reduction using PCA. Result showed that the visualized EEG feature
distributions of watching 2D and 3D at early stage were different then the
distributions became similar according as the flow of watching time. From that,
greater 3D visual cognitive load at early stage became to stabilize according as the
flow of watching time.
In future works, we will perform additional experiments for long time watching
such as longer that one hour. From that, the cognitive tendency of human brain when
watching 3D display will be more accurately analyzed. Also, we will find
connectivity between EEG features and autonomic nervous system features.

Acknowledgments. This work was supported by the Global Frontier R&D

Program on <Human-centered Interaction for Coexistence> funded by the National
Research Foundation of Korea grant funded by the Korean Government


1. Lee, E.C., Heo, H., Park, K.R.: The Comparative Measurements of Eyestrain Caused by 2D
and 3D Displays. In: IEEE Transactions on Consumer Electronics. vol. 56, pp. 1677--1683
2. Kim, Y.J., Lee, E.C.: EEG Based Comparative measurement of Visual Fatigue Caused by
2D and 3D Displays. In: Communications in Computer and Information Science. vol. 174,
pp. 289--292 (2011)
3. Information on Bioelectromagnetism Education and Research,
4. Bruce, M.: Principal Component Analysis in Linear Systems: Controllability, Observability,
and Model Reduction. In: IEEE Transactions on Automatic Control. vol. 26, pp. 17--32
5. Friston, K.J., Frith, C.D., Liddle, P.F., Frackowiak, R.S.J.: Functional Connectivity: The
Principal-Component Analysis of Large (PET) Data Sets. In: Cerebral Blood Flow and
Metabolism, vol. 13, pp. 5--14 (1993)

Copyright 2015 SERSC 75