Eye tracking is a technique whereby an individual’s eye movements are measured so that the researcher knows both where a person is looking at any given time and the sequence in which their eyes are shifting from one location to another (Poole and Ball 2006) . Over the past decade, eye tracking with highly specialized eye wear equipment has been used to record detailed and accurate eye movements and visual direction in many psychology studies (Gilland 2008) . Kiefer and his colleagues from ETHZ studied self-localization and human wayfinding using location-aware mobile eye tracking (Kiefer, Straub et al. 2012) , which looked at the gazing patterns on the map to determine the participants’ critical decision points. Analysis and recording of eye movements has also been an important tool in the investigation of the driver’s visual awareness and driver behaviours in dynamic driving situations, particularly a driver’s spatial cognition and fixation (Falkmer, Dahlman et al. 2008; Dukic and Broberg 2012) . Eye tracking enables the researcher to collect data relating to cognitive processes employed while undertaking any particular task such as turning, these processes may include the order and length of time a viewer directs gaze at any particular object in a visual scene (Falkmer, Dahlman et al. 2008) . Fixation is a central aspect of eye tracking analyses, and may be defined as the length of time the eye ceases movement and remains set on any particular focal point (Green, 2002; Zwahlen & Schnell, 1998). Fixation is relatively “stationary” eye behaviour, which allows eyes to focus their gaze on the objects being looked at, and to extract this information (Yang, McDonald et al. 2012) . It is during these fractions of a second that the brain is able to receive visual information which has been acquired from the focal point (Gilland 2008) . Therefore, this method utilizes eye tracking to gain an accurate picture of a driver’s visual pattern and can be used to evaluate driving behaviours.

It can be comparably low-cost and ecologically valid to assess driving behaviours using Global Positioning System (GPS) tracking. GPS provides a feasible way to continuously measure the position, velocity and acceleration of a vehicle under typical driving conditions. In previous work, GPS or the combination of GPS plus video recording can provide a means to assess driving behaviours by tracking vehicle movements (Porter and Whitton 2002; Grengs, Wang et al. 2008; Naito, Miyajima et al. 2009; Cruz, Macías et al. 2013; Mudgal, Hallmark et al. 2014) . A Multi-GNSS (Global Navigation Satellite System) receiver is the system able to calculate position, velocity and time by receiving the satellite signals broadcasted from multiple satellite navigation systems. Currently, there are a number of GNSS systems under operating including GPS, GLONASS, Galileo, QZSS and Beidou (Kubo, Hou et al. 2014; Noomwongs, Thitipatanapong et al. 2014) . Using multiple satellite systems can achieve high position accuracy with increased number of satellites compared to GPS-only positioning, particularly in harsh environments (urban canyon etc.) where the GPS-only positioning becomes difficult (Kubo, Hou et al. 2014; Noomwongs, Thitipatanapong et al. 2014) . Due to the nature of satellite signals, GNSS raw data contains noises. The accuracy of GNSS data depends on many factors, the position of the satellites at the time the data was recorded, errors in satellite clocks and orbits, the trips through the layers of the atmosphere, and many other sources contribute inaccuracies to satellite signals by the time they reach a receiver (Sun, Odolinski et al. 2014) . For such consideration, the relative GNSS techniques can be introduced to improve the accuracy by minimising the effect of each error source. The most complex relative technique is RTK (real-time kinematic) GNSS, which uses a known position of a base station to computer the moving receiver position. By having the base station over a known position, the errors of atmospheric effects can be estimated and referenced to the observed position of the receiver. This allows for relative positioning accuracies of sub-meter level and even down to decimetre level. From there, vehicle speed, acceleration/deceleration and lateral position data can be generated from the trajectory positions for driving behaviour assessment.

Where and how long and how often the driver gazed at during driving are indicative of the visual perception strategy for the drivers to keep the car in an optimal lane position, in fact, cognitive resources involving visuospatial and motor coordination are required for driving manoeuvres. To investigate the visual-motor coordination in driving, the eye movement tracking can be integrated with the vehicle movement tracking by synchronising the two datasets. Figure 1 below illustrates the structural framework of such approach.

The eye tracking data including the fixation objects and duration of visual searching behaviour can be geocoded using vehicle position reference, and linked to the parameters on vehicle movement trajectory, such as speed, acceleration/deceleration, vehicle head angle and vehicle lateral position to the road centreline (Table 1). Such data including spatial and temporal attributes for both visual pattern and vehicle movement, while the latter reflects how the driver controlled the steering wheel and brake and accelerator pedals. The geocoding and synchronising can be accomplished in GIS (Geographic Information Systems) environment; the outcome dataset is presented in table 1. The (x, y) coordinates of the visual-motor behaviour data can be overlaid with other environment and transport information in GIS, by analysing spatial-temporal patterns of the synchronised eye fixation and vehicle movement data, the characteristics of the driver’s visual-motor can be investigated in depth. A prototype implementation of the framework in next section presents the procedure and methods to further investigate the older drivers’ visualmotor coordination of U-turn manoeuvre at a roundabout.

Parameters of Eye movement n ) ito dn a o r c uD (se c i f f a r T 1 2 1 2 2 2 ) h / m k ( d e e p

S 7.02 6.85 6.67 6.46 6.39 6.39

Previous studies found that older drivers were over-represented in angle collisions, crashes at intersections, turning and changing lanes (McGwin Jr 1999; Clarke, Ward et al. 2010; Marmeleira, Ferreira et al. 2012) . Those evidences indicate that the age-related decline in particular function leads to unsafe driving (Dobbs, Heller et al. 1998; McKnight 1999; Fancello, Pinna et al. 2013; Wood, Horswill et al. 2013) . Even so, not all older drivers are unsafe, the statistics don’t reflect driving abilities of individuals; age itself, isn’t the predictor of the fitness of driving (Anstey, Wood et al. 2005) . While there is a strong emphasis around the world for older adults to maintain their mobility for as long as possible, the challenge is to develop appropriate evaluation methods to identify those older drivers with hazardous driving behaviours and to provide intervention as early as possible (Lee 2003) . Worldwide, the ageing population has brought the issue of older drivers into a sharper focus. In this case study, we set up on-road driving experiment in older drivers in order to investigate their visual-motor coordination behaviour.

Three female and two male older drivers aged from 60 to 79 (mean=67, SD =7.2) were recruited from local community. The eligibility of participation also includes: holding a valid Australian driver license and having an insured vehicle, driving at least 3-4 times a week, and having no mental and physical issues affecting driving. Before the assessment, all subjects provided informed consent for participation in compliance with ethics requirements from the Curtin University HREC (Human Research Ethics Committee), followed by an eye acuity check and the Mini-Mental State Examination (MMSE) to ensure their basic fitness for on-road driving. A miniquestionnaire survey on demographics and driving habits was also conducted prior to the assessment.

The study area was chosen around the campus of Curtin University in Perth, approximately 1.5km radius distance from the Curtin GNSS (Global Navigation Satellite System) base station. This enables the cost-effective RTK postprocessing techniques for precise vehicle movement data using the base station reference data. The primary purpose of the on-road driving assessment was to track the driver’s visual-motor coordination at the complex roundabout manoeuvres. This was achieved by simultaneously recording the driver’s eye fixation and the vehicle movement trajectory.

During on-road driving test, participants were asked to drive their own cars with eye tracker mounted on their heads (Figure 2). A 16-point calibration procedure was carried GNSS receivers out prior to the on road test. Eye-tracker: A head mounted Arrington Viewpoint TM eye tracker was used to measure the eye movements of participants in 60Hz.

Data collected including number of fixations and fixation durations, a proxy to driver’s attention and Eye fixation reaction to external stimuli from the environment are captured by the eye-tracker system. Corrected vision was required and the eye tracker can be worn with glasses when necessary. These processes may include the order and length of time a driver directs gaze at any particular object in a visual scene, as well as the visual patterns the driver utilizes while performing any particular driving task (Falkmer, Dahlman et al. 2008) .

The sequential video clips of drivers view and eye fixation objects (the green dots) in Figure 3 show the visual pattern of the driver during the roundabout: from left to right and top to bottom, the driver looked frequently Figure 2: Equipment and on-road driving setting at the passing cars and edge line of the curves in order to avoid the hazards and coordinate the vehicle positions.

A pair of Trimble R10 GNSS receivers was mounted on each participant’s car roof to record the vehicle movement trajectory (Figure 2). The receivers are able to track multi-GNSS systems beyond GPS only approach, this is essential for recording vehicle movement in urban area when the precise vehicle positioning and smoothest trajectory are required. The tracking configuration of Trimble R10 receivers was setup at 10Hz in order to record the car positions at every 0.1 second. Real-time kinematic (RTK) postprocessing technique was used to achieve centimetre to decimetre level horizontal accuracy by minimizing the effect of error sources transmitted between the satellites and GNSS receivers (Sun, Odolinski et al. 2014) . The post-processed data was then mapped to calculate the lateral position, speed and acceleration. Figure 4 shows one diver’s speed control (left) and lane keeping (right) at the roundabout manoeuvre: slowing down to stop when approaching the roundabout, keeping moderate speed in the roundabout and accelerating to exit the roundabout; the perpendicular red lines between the car position to the road centreline demonstrate the lane deviations of the manoeuvre, in the case of this driver, the exit of the roundabout gave more lane deviation, followed by the stage when entering into the roundabout (west side of the roundabout). Further analysis based on the speed and lane keeping data between the drivers can be seen in figure 6.

All the five older drivers’ manoeuvres were analysed in GIS platform, the core dataset combining both the eye movement behaviours and speed control and lane deviation was overlaid with road centrelines and orthoimagery in order to visualise the spatial-temporal patterns of visual-motor coordination in individuals. As shown in figure 5 above, the driver smoothly decelerated when approaching the roundabout from 3140km/h to 21-30km/h then to 11-20km/h to prepare to enter the roundabout and steadily accelerated through the roundabout and while exiting the roundabout to 41-48km/h. Eye tracking data shows that the driver looked right and looked into the curve while preparing to enter the roundabout, specifically at line markings, and other objects. While travelling through the roundabout the participant primarily looked into the curve at the inner rear view mirror, the right rear view mirror, trees/shrubs, and line markings. When exiting the roundabout the participant looked right at the right rear view mirror, looked straight ahead at tree or shrub, looked into the curve at trees/shrubs and line markings, and looked left at trees/shrubs. The participant showed adequate preparation while entering the roundabout and active scanning patterns while travelling through the roundabout, with the majority of fixations being traffic relevant objects. The visualisation of the spatial-temporal patterns of visual-motor coordination would give more insight into the understanding of how the drivers used visual information to control their physical movement in particular manoeuvres.

Figure 6 below shows the variations of visual-motor coordination through roundabout manoeuvres among 5 older drivers. The left graph shows the duration and numbers of fixations within individual drivers; and the right plots the mean and std of speed control and lane deviation. As can been interpreted, Driver4 had the most frequent eye fixation and longest duration at curves, presented the lowest lane deviation, and slight higher mean speed. Driver3 had the least frequent eye fixation overall but average eye fixation and duration at curves, this participant demonstrated the highest lane deviation but the lowest std (standard deviation) value of lane deviation, and the slowest mean speed. Other drivers also used different strategies of visual-motor coordination when negotiating roundabouts. The eye fixation at curves seems associated with lane keeping and speed control, but there might be other conditions leading to the driving performance at roundabouts.

The descriptive statistical analysis in this case study tells that older drivers with better visual searching strategy achieved slightly less lane deviation and higher mean speed at roundabout manoeuvres. The visualisation of spatialtemporal patterns of visual-motor coordination when entering, passing, exiting roundabout gives more insight into the understanding of how the drivers used visual information to control their physical movement in particular manoeuvres. The findings of this study indicated that those individuals at inflated risk of road crashes could be identified using the combination of eye tracking and vehicle movement tracking in order to detect the detailed behaviours which can be hardly observed by other methods, such as observation by instructors, clinical assessments or driving simulations, etc.

60 50 In summary, this case study showed that combining eye tracking and precise vehicle movement tracking can detect the variation of the visual-motor coordination in older drivers. This study has also presented a cost-effective and ecologically valid approach in driving behaviour assessment, which can be used in studies with larger sample size in the future. Further advanced statistical analysis method, such as DFA (discriminant function analysis), can be used to predict variables affecting visual-motor coordination in older adults with high discriminant validity.

The authors would like to thank the GNSS Research Centre, Curtin University for providing base station reference data and the, eye tracking analysis team from the School of Occupational Therapy and Social Work, Curtin University, for analysing eye tracking data.