Economical Visual Attention Test for Elderly Drivers
EEconomical Visual Attention Test for Elderly Drivers
Akinari Onishi a,b, ∗ a Chiba University, 1-33, Yayoicho, Inage-ku, Chiba-shi, Chiba, Japan b National Institute of Technology, Kagawa College, 551 Kohda, Takuma-cho, Mitoyo-shi,Kagawa, Japan
Abstract
Traffic accidents involving elderly drivers are an issue in a super-aging society.A quick and low-cost aptitude test is required to reduce the number of trafficaccidents. This study proposed an oddball-serial visual search task that as-sesses the individual’s performance by his or her responses to the presence ofcued stimuli on the screen. Task difficulty varied by changing the number ofsimultaneous stimuli; Accordingly, low performers were detected. In addition,performance correlated with age. This implies that individual characteristicsrelated to driving performance that decline with age can be detected by theproposed task. Since the task requires low-cost devices (computer and responsebutton), it is feasible for use as a quick and low-cost aptitude test for elderlydrivers.
Keywords:
Elderly drivers, aptitude test, oddball, visual search,EEG
1. Introduction
It was estimated in 2001 that approximately a quarter of the people in theOrganisation for Economic Co-operation and Development (OCED) Membercountries would be older than 65 years by 2050 [1]. Japan became a super-aging ∗ Corresponding author
Email address: [email protected] (Akinari Onishi )
URL: http://onishi.starfree.jp/ (Akinari Onishi )
Preprint submitted to Transportation Research Part F: Traffic Psychology and BehaviourAugust 12, 2020 a r X i v : . [ c s . H C ] A ug ociety in 2007, and numerous associated issues have arisen [2]. One issue in asuper-aging society is traffic accidents involving elderly drivers [3].Elderly drivers have unique driving characteristics. For example, elderlydrivers are relatively likely to crash at intersections [4]. Such crashes are relatedto failures of attention [5]. In fact, traffic accidents involving elderly driversare related to their physical, motor, sensory, and cognitive factors. Since vi-sual function changes with aging, various studies of vision and vision-relatedcognitive functions have been conducted. The correlation between vision testsand crash records was weak, while attention scores were correlated with crashrisk and driving performance [6]. To prevent serious traffic accidents involvingelderly drivers, assessments of visual attention are essential. Such assessmentswould enable elderly drivers to reduce risks by, for example, installing driverassistance systems [7], or by training physical or cognitive functions [8].Various driving assessment systems utilize driving simulators (DSs). MostDS utilize driving scenes that contain events, such as a pedestrian crossing infront of the driving vehicle. A study employed a DS to investigate the rela-tionship between mental workload and age [9]. In addition, some DS tests havebeen proposed to measure the characteristics of elderly drivers. For example,one DS test measures the range of the peripheral visual area that is relatedto the cognitive function named the useful field of view (UFOV) [3]. Further-more, a DS test has been reported that clarifies the characteristics of the elderlydriver’s response by recording their palmar sweating response, galvanic skin re-sponse, and driving skills [10]. Therefore, DS tests are helpful for revealing thecharacteristics of elderly drivers in detail. However, these are not suitable aswidely applied aptitude tests for elderly drivers because of test-related costs andhigh time demands. For example, most DSs are expensive and physically large.Moreover scenario-based tests should be repeated many times in order to permitrobust statistical analysis. In a super-aging society, many elderly drivers requireaptitude tests. Thus, tests should be able to be conducted given limited time,space, and cost. Accordingly, minimal psychological tests should be considered.Numerous psychological and physiological measures have been used to eval-2ate the driving performance of elderly drivers [11]. Cognitive test scores (e.g.,Mini-Mental State Exam) have been shown to correlate with in-traffic drivingperformance scores [12]. Moreover, driving relies on vision, including visualattention [13], and visual cognitive function is influenced by aging [14, 15].One well-studied visual function score is the UFOV [16, 17]. The UFOV de-clines with age [18, 19, 20] and is highly predictive of the likelihood of accidentsamong elderly drivers [21]. The UFOV is measured using a salient stimulus inperipheral vision, in a form of pop-up task. However, dangers hidden in actualscenes are not salient. In addition to those subjective evaluation, performanceshould be estimated by an objective evaluation using physiological signals suchas electroencephalography (EEG) [22].This study developed a low-cost, rapid aptitude test for elderly drivers thatcombined a visual search task with an oddball task. Since psychological tests aremore economical than DS tests, this study focused on a simple psychological testthat required only a computer and a response button. The difficulty of the taskwas varied in three conditions to clarify each subject’s performance. Young andelderly participants were recruited to determine individual differences in eachgroup, in addition to differences between groups.
2. Materials and methods
Fifteen elderly participants (68.7 ± ± In this study, a participant was asked to respond to the cued stimulus bypushing a button. The task was an oddball-serial visual search (OSVS). The3 able 1: Profiles of participants.
ID Age Gender Handedness Vision Correction Driver’s licence (experience)y1 31 M Right Spectacles Yes (13 years)y2 22 M Right Contact lenses Yes (3 years)y3 23 M Right Spectacles Yes (5 years)y4 22 M Right No Yes (4 years)y5 23 M Right Contact lenses Yes (4 years)y6 21 M Right Contact lenses Yes (3 years)y7 24 M Right Contact lenses Yes (4 years)y8 22 M Right Contact lenses Noy9 21 F Right Contact lenses Noy10 22 F Right Spectacles Yes (3 years)e1 69 M Right Spectacles Yes (42 years)e2 65 M Right No Yes (28 years)e3 70 M Right Spectacles Yes (51 years)e4 70 M Right No Yes (29 years)e5 66 F Right Spectacles Yes (34 years)e6 67 F Right Spectacles Yes (46 years)e7 71 F Right No Yes (53 years)e8 69 M Right Spectacles Yes (46 years)e9 76 M Right No Yes (55 years)e10 69 M Both Spectacles Yes (36 years)e11 72 M Right Spectacles Yes (54 years)e12 67 F Right Spectacles Yes (51 years)e13 65 F Right No Yes (26 years)e14 65 F Right No Yes (47 years)e15 70 F Right Spectacles Yes (37 years)4 ress the button whenthe following stimulus occurs m s ~ m s ~ m s CueNontarget TargetNontarget T i m e … Figure 1: Example of the oddball-serial visual search task. First, the target stimulus wascued. After 500 ms, three stimuli were simultaneously presented in a horizontal orientation inthe middle of the screen (P3 condition). New sets of stimuli were continuously presented every1000 to 1800 ms. The participant pushed a button using his or her right-hand as quickly aspossible (right-hand condition) if the target stimulus was present anywhere on the screen. Ifno target stimulus appeared on the screen (nontarget stimulus), the participant was instructedto ignore the stimulus. The task was repeated during a session. participant was seated 60 cm in front of the monitor (E178FPc, DELL), holdinga button box. This study employed 8 circular stimuli; part of the circle in oneof 8 directions was missing. The stimuli are similar to Landolt Cs. Figure 1demonstrates an example of a trial. First, the participant remembered a cuedtarget stimulus. Then, multiple stimuli were horizontally presented simultane-ously for 500 ms. If the target stimulus appeared, the subject was requiredto press the button as quickly as possible. After a response was made, thenext stimuli were presented after a 500 to 1300 ms delay; the stimulus-onsetasynchrony was 1000 to 1800 ms. 5 timuliTrialsSets CueTrial 1 Trial 2 Trial 3P1(R1) P1(L2)P3(L2)P3(L1) P5(L1)N N N N N N TT N …… T:Target, N:NontargetCondition(Hand, Set ID)
Figure 2: Experimental procedure. Each trial began with a cue. After the cue, 40 stimuli weredisplayed pseudo-randomly, in succession. Two types of stimuli were provided: target stimuli(T), which required a response, and nontarget stimuli (N), which did not require a response.Three blocks comprised a set of experimental trials. Each set consisted of one experimentalcondition, namely the stimulus condition (P1, P3, P4), hand condition (R: right, L: left),and set ID (1, 2). Each stimulus condition contains 480 stimuli. For example, P3 conditioncomprise of 4 sets, namely P3(R1), P3(R2), P3(L1), and P3(L2) sets. Those 4 sets (handsand twice) contain 3 trials, and each trial has 40 stimuli, respectively.
The experimental design is shown in Fig. 2. There were 40 stimulus presen-tation in each block of trials, in which target stimuli appeared 8 times. That is,the probability that the target stimulus appeared was 1/5. Three blocks of trialswere conducted in an experimental sequence. The sequence was conducted withright and left hand conditions. Three stimulus conditions were prepared; one(P1), three (P3), or five (P5) different stimuli were presented simultaneously(see Fig. 3). Furthermore, all conditions were repeated twice. There were 480trials for each stimulus condition (40 stimuli × × × EEG signals were recorded during the experiment. This study used thePolymate mini AP108 (Miyuki Giken Co., Ltd, Tokyo, Japan), a portable EEG6 b)(a)(c)
Figure 3: Stimulus conditions. P1 condition (a) a single circle with a missing segment,displayed at the center of the screen. P3 (b) and P5 (c) conditions show three and five circleshorizontally along the vertical center of the screen, respectively. amplifier with active electrodes. Electrodes were located at Fpz, Fz, Cz, Pz, andOz in accordance with the international 10-20 system. The reference and groundelectrodes were placed at A2 and Afz, respectively. The sampling rate was 500Hz. Filters applied consisted of a hardware low-pass filter (cut-off frequency:30 Hz), high-pass filter (time-constant: 1.5 s), and notch filter (50 Hz). Thesensitivity of the amplifier was 20 µ V. 7 mm mm Figure 4: Location and size of the stimuli on the monitor.
This study focused on visualizing the raw data to confirm the presence ofindividual differences among the subjects. Task performance was analyzed us-ing the frequency of true-positive (TP), true-negative (TN), false-positive (FP),and false-negative (FN) responses, in addition to accuracy, precision, and sensi-tivity. Further, descriptive statistics were used to summarize the results of theexperiments. Medians were used as the measure of central tendency of the data,because of the presence of ceiling effects. Correlation analysis was also used toclarify the relationships among profiles of participants and performance.8 . Results
To clarify each individual’s performance, all participants conducted the OSVStask for each stimulus condition (P1, P3, P5). Figure 5 represents the numberof TP, TN, FP, and FN responses during the task.The median TP of the young participants was 96 in P1, 89 in P3, and 64.5in P5 (TP - Young). Median TP of the elderly participants was 96 in P1, 75in P3, and 50 in P5 (TP - Elder). Large variance in TP can be seen in P3 andP5 for elderly participants, and in P5 for young participants. Participant y2exhibited the lowest TP in P3, at 52 such responses, which was far from themedian. The TP of participants y2 and y4 in P5 was 52 and 49, respectively.Participants e3, e4, and e9 produced 37, 39, and 36 such responses, respectively.TP during P5 was low for most elderly participants.The variance in TN was small across conditions. TN for young participantswas 383, 380.5, and 377.5 (TN - young), while TN for elderly participants was382, 377, and 378 (TN - Elder) for P1, P3, and P5 conditions, respectively. TNof participant e7 in P3 condition was 353, which was lower than the median.The line chart shows that the FP and FN plots are essentially invertedversions of the TN and TP plots, respectively.
Accuracy, precision, and sensitivity were calculated, as indicated in Fig. 6.Median accuracy of the young participants was 99.8, 98.2, and 92.5%, while thatof elderly participants was 99.6, 94.4, and 87.9% for P1, P3, and P5 conditions,respectively. Participant y2 exhibited the lowest accuracy among young subjectsfor all stimulus conditions. Participants e3 and e7 exhibited decreases in the P3condition. Most elderly subjects demonstrated low accuracy in the P5 condition.The precision values reflected individual differences among the elderly par-ticipants. Median precision of the young participants was 99.0, 96.3, and 91.9%,9
P - Young TN - Young FP - Young FN- YoungTP - Elderly TN - Elderly FP - Elderly FN- Elderly
Figure 5: Individual’s TP, TN, FP, and FN responses in each stimulus condition (P1, P3, P5). while that of elderly participants was 98.0, 90.6, and 87.2% for P1, P3, and P5conditions, respectively. Some elderly participants demonstrated high precisionin P3 and P5 conditions, while participants e2, e3, and e10 exhibited markeddecreases in P3 and P5 conditions. These tendencies were not observed in ac-curacy and sensitivity. Participant e7 exhibited the lowest precision in P3, butapproximately median precision in P5.Although the shape of the trend in median sensitivity was similar to that ofaccuracy, the trends for individual participants differed between accuracy andsensitivity. For example, accuracy of participant e4 was close to the median,while sensitivity was among the worst.10 ccuracy - Young Precision - Young Sensitivity - Young Response time - YoungAccuracy - Elderly Precision - Elderly Sensitivity - Elderly Response time - Elderly
Figure 6: Individual accuracy, precision, sensitivity, and response time for each stimuluscondition.
Responses time are presented in Fig. 6. The median response times of theyoung participants were 0.39, 0.56, and 0.63 s, while those of the elderly par-ticipants were 0.50, 0.66, and 0.68 for P1, P3, and P5 conditions, respectively.Two types of tendency can be seen: (1) response time gradually increased as thecondition changed from P1 to P3, and (2) response time saturated at P3 anda similar response time was observed in P5. Most young participants exhibitedthe former tendency, while most elderly participants demonstrated the latter.Participant y2 exhibited the latter type. The peak response time of participante9 occurred at P3. 11
RP amplitude - Young ERP latency - YoungERP amplitude - Elderly ERP latency - Elderly
Figure 7: Individual ERP amplitude and ERP latency for each stimulus condition.
ERP amplitude and EPR latency are shown in Fig. 7. Large differences inmedian ERP amplitude cannot be seen across stimulus conditions because ofits large variance. Median EPR amplitude for young participants was 9.8, 8.3,and 6.6 µ V, while that of elderly participants was 4.8, 2.1, and 3.1 µ V for P1,P3, and P5 conditions, respectively.ERP latency tended to increase as the stimulus condition moved from P1to P5. Median ERP latency for young participants was 0.52, 0.63, and 0.66 s,while that of elderly participants was 0.58, 0.65, and 0.70 s, respectively.
The Friedman test was applied to the results shown in Fig. 5, 6, and 7, toassess the differences among stimulus conditions for each group. The results12f the statistical tests are summarized in Table. 2. TP differed significantlyamong all stimulus conditions. Significant differences can be seen between P1and P5 for all indices, except for ERP amplitude and ERP latency of youngparticipants, and ERP amplitude for elderly participants. P1 and P3 conditionsalso revealed significant differences between all but TN, FP, and ERP amplitudefor young participants, and all but ERP amplitude and ERP latency for elderlyparticipants. Significant differences between P3 and P5 were confirmed for TP,FN, accuracy, precision, sensitivity, and reaction time for young participants,and TP, TN, accuracy, and sensitivity for elderly participants.
Table 2: Friedman test.
Condition Young ElderlyP1-P3 P1-P5 P3-P5 P1-P3 P1-P5 P3-P5TP * * * ** ** **TN n.s. * n.s. * ** n.s.FP n.s. * n.s. * ** n.s.FN * * * ** ** **Accuracy * * * ** ** **Precision * * * ** ** n.s.Sensitivity * * * * * *Reaction time * * * * * n.s.ERP amplitude n.s. n.s. n.s. n.s. n.s. n.s.ERP latency * n.s. n.s. n.s. * n.s. *: p < . , **: p < . , ***: p < . . .3. Correlations Correlation analysis was applied to results of Fig. 5, 6, and 7, in addition tothe profile of participants (Table. 1). Correlations among age, TP, TN, FP, andFN are shown in Table. 3. Significant negative correlations were present betweenage and TP in P3 and P5 conditions. Consistent with the inverted relationshipbetween TN and FN was in Fig. 5, positive correlations were observed betweenage and FN in P3 and P5 conditions. There were no significant correlationsbetween age and FP nor between age and TN.
Table 3: Correlations among age, TP, TN, FP, and FN.
Condition TP FNP1 P3 P5 P1 P3 P5Age n.s. -0.43 -0.58 n.s. 0.43 0.58Condition FP TNP1 P3 P5 P1 P3 P5Age n.s. n.s. n.s. n.s. n.s. n.s.
Correlation coefficient (value) is listed if the correlation was significant.
Table. 4 lists correlations among age, accuracy, precision, and sensitivity.Significant negative correlations were observed between age and accuracy (P3,P5), between age and precision (P5), and between age and sensitivity (P3, P5).
Table 4: Correlations among age, accuracy, precision, and sensitivity.
Condition Accuracy Precision SensitivityP1 P3 P5 P1 P3 P5 P1 P3 P5Age n.s. -0.47 -0.64 n.s. n.s. -0.43 n.s. -0.43 -0.58
Correlations coefficient (value) is listed if the correlation was significant.
Table. 5 presents correlations between age and reaction time. Correlationswere significant for P1 and P3 conditions.14 able 5: Correlations between age and reaction time.
Condition Reaction timeP1 P3 P5Age 0.51 0.64 n.s.
Correlation coefficient (value) is listed if the correlation was significant.
Table. 6 presents correlations among age, ERP amplitude, and ERP latency.Only age and ERP amplitude in the P1 condition were significantly negativelycorrelated.
Table 6: Correlations among age, ERP amplitude, and ERP latency.
Condition ERP amplitude ERP latencyP1 P3 P5 P1 P3 P5Age -0.51 n.s. n.s. n.s. n.s. n.s.
Correlation coefficient (value) is listed if the correlation was significant.
Table. 7 summarizes the correlations among TN (P1), FP (P1), accuracy(P1), precision (P1), ERP amplitude, and ERP latency. TN in the P1 conditioncorrelated positively with ERP amplitude in P1 and P3 conditions. FP in theP1 condition and ERP amplitude (P1, P3) were negatively correlated. Therewas a significant correlation between accuracy in the P1 condition and ERPamplitude in the P1 condition. Precision in the P1 condition also correlatedpositively with ERP amplitude in P1 and P3 conditions.15 able 7: Correlations among TN (P1), FP (P1), accuracy (P1), precision (P1), ERP ampli-tude, and ERP latency.
Condition ERP amplitude ERP latencyP1 P3 P5 P1 P3 P5TN (P1) 0.51 0.44 n.s. n.s. n.s. n.s.FP (P1) -0.51 -0.44 n.s. n.s. n.s. n.s.Accuracy (P1) 0.45 n.s. n.s. n.s. n.s. n.s.Precision (P1) 0.51 0.43 n.s. n.s. n.s. n.s.
Correlation coefficient (value) is listed if the correlation was significant. Data notshown in this table, such as TN(P3) and TN(P5) conditions in addition to TP, FN,and sensitivity, did not exhibit any significant correlations with either ERP amplitudeor ERP latency. . Discussion The OSVS task was proposed and evaluated to develop a low-cost, rapidaptitude test for elderly drivers. During the task, the number of stimuli wasswitched to 1, 3, and 5 in P1, P3, and P5 conditions, respectively. Reactionswere measured by button-press and EEG. As a result, poor performers wereconfirmed in TP, FN, accuracy, precision, and response time. Furthermore, TP,FN, accuracy, precision, and response time in the P3 condition were correlatedwith age. These results imply that performance measured by button-press re-sponses may be indicative of driving performance among elderly drivers. Theprototype test device without EEG was cheap, because it consisted of only acomputer and button-response box. In addition, the task duration was some-what extended so as to conduct a rigorous psychological experiment, but theduration could be shortened. Therefore, the OSVS task may be suitable as alow-cost, rapid aptitude test for elderly drivers.This study showed that in the P3 condition, some performance measurescorrelated with age. The P3 condition of the OSVS task is a type of selectiveattention task. As mentioned in the introduction, age is related to decline invisual cognitive functions, attention, and UFOV. Therefore, the results of thisstudy are consistent with those of previous studies. Some young participantsalso exhibited poor performance in the P3 condition. The results imply thatsome young drivers may have poor attention-related performance, which couldbe detected by the OSVS task. Indeed, young drivers are relatively likely tocause serious driving accidents. Young drivers often believe that their drivingperformance is high; as such, they tend to overestimate their own skills. Incontrast, a study showed that young drivers are not optimistic [23]. Youngdrivers tend to cause accidents in some situations, such as late at night [24, 25].Therefore, aptitude tests or training for young drivers could reduce such trafficaccidents. That is, the OSVS task may contribute to reducing traffic accidentsamong younger drivers as well as among elderly drivers.Although the OSVS test was intended as an aptitude test that is used when17btaining or renewing driver’s licences, the test could be used in other applica-tions. For example, it could be employed as a self-assessment when introducinga driving assistance system [7] or automated driving in future [26, 27]. Further-more, it would be interesting if the training effect of the OSVS test was clarified.In this respect, training methods for elderly drivers have been proposed. Forexample, a trail making game system has been proposed for training [8]. Ad-ditionally, a study proposed a video-based trail making test for young drivers[28]. OSVS has the potential to be extended as a training tool for young andelderly drivers.EEG amplitude correlated with responses in P1, but not those in P3 or P5.That is, EEG amplitude was not associated with selective attention. The OSVStask with the P3 or P5 conditions is a type of selective attention task, while theP1 condition represents a sustained attention task. Indeed, the oddball task isused for measuring sustained attention [29].In future studies, the correlation between OSVS scores and the numberof traffic accidents should be identified. Previous studies reported that thecorrelation between cognitive function scores and traffic accidents was low [30,31]. In contrast, UFOV test outcomes were correlated with the number oftraffic accidents, as mentioned in the introduction. OSVS scores, which canbe obtained without the need for expensive devices, were not compared withthe number of traffic accidents. Accordingly, OSVS task scores and drivingsimulator tests should be compared. Such a comparison is required to determinethe threshold of OSVS scores for the aptitude test. Additionally, a shortenedOSVS task should be evaluated to establish a low-cost, rapid aptitude test.
5. Conclusion
This study developed a low-cost, rapid aptitude test for elderly drivers. Sincepsychological tests are more economical than are DS tests, this study focusedon a simple psychological test that required only a computer and a responsebutton. This study proposed a new aptitude test that combined a sequential18isual search task with an oddball task. The number of stimuli varied in threeconditions to clarify individual differences in performance. Poor performers wereconfirmed in terms of TP, FN, accuracy, precision, and response time. Further,those scores were correlated with age when three stimuli were simultaneouslypresented. The proposed task may contribute to developing a low-cost, rapidaptitude test for elderly drivers.
6. Acknowledgements
This study was supported in part by the Mitsui Sumitomo Insurance WelfareFoundation.