Reliability of GPS based traffic data: an experimental evaluation

Working papers in transport, tourism, information technology and microdata
analysis
Reliability of GPS based traffic data: an experimental evaluation
Author 1: Xiaoyun Zhao
Author 2: Kenneth Carling
Author 3: Johan Håkansson
Editor: Hasan Fleyeh
Nr: 2014:17
Working papers in transport, tourism, information technology and microdata analysis
ISSN: 1650-5581
© Authors
Reliability of GPS based traffic data: an experimental evaluation
Xiaoyun Zhao, Kenneth Carling, Johan Håkansson
Abstract
GPS tracking of mobile objects provides spatial and temporal data for a broad range of
applications including traffic management and control, transportation routing and planning.
Previous transport research has focused on GPS tracking data as an appealing alternative to travel
diaries. Moreover, the GPS based data are gradually becoming a cornerstone for real-time traffic
management. Tracking data of vehicles from GPS devices are however susceptible to
measurement errors – a neglected issue in transport research. By conducting a randomized
experiment, we assess the reliability of GPS based traffic data on geographical position, velocity,
and altitude for three types of vehicles; bike, car, and bus. We find the geographical positioning
reliable, but with an error greater than postulated by the manufacturer and a non-negligible risk
for aberrant positioning. Velocity is slightly underestimated, whereas altitude measurements are
unreliable.
Key words: GPS tracking device, reliability, transportation, road network
1. Introduction
Global Positioning System (GPS) is a Global Navigation Satellite System (GNSS) for geopositioning. The availability and usability of GPS devices in geo-positioning and tracking mobile
objects has grown enormously in the past decades and is still increasing. The GPS has emerged
for civilian use in the 1990s as the space geodetic technique being accurate and economical
(Zumberge et al., 1995). In their review, Theiss et al. (2005) identified a wide range of
applications of GPS tracking data including timing, logistics, traffic management, and weather
forecasting and concluded that it will change the way that companies and organization run their
business.
GPS tracking technologies have extensively been applied in transportation studies, in particular
for studying the routes of motorized vehicles (Zito et al., 1995; Quiroga and Bullock, 1998;
Murakami and Wagner, 1999). For instance, Schönfelder (2002) presented an innovative
approach to collect GPS longitudinal travel behavior data on humans and described the
complexity of their daily life with the interaction between periodicity and variability.
GPS is also applied to study the travel pattern and mobility prediction (Ashbrook et al. 2002 and
2003,). For instance, Jia et al. (2012) confirmed the scaling property and identified the Levy
flight characteristic of human mobility by using the GPS tracking data of car movements. GPS
data is also applied in environment control. For instance, Carling et al. (2013) and Jia et al. (2013)

Xiaoyun Zhao is a PhD-student in Micro-data analysis and corresponding author: xzh@du.se, phone: +46 23778509. Kenneth Carling is a professor in Statistics and Johan Håkansson is a professor in Geography. All are at the
School of Technology and Business Studies at Dalarna University, Sweden
studied the induced pollutant emissions of CO2 from car movements by using a GPS tracking
data of car movements.
Even though GPS tracking data opens up for interesting applications, gathering information of
spatiotemporal mobility by GPS is subject to critical reflections. Leduc (2008) examined recent
developments in road traffic data collection and discussed the potentials and bottlenecks related
to new GPS technologies. Moreover, Van der Spek et al. (2009) concluded that GPS offers a
widely useable instrument to collect invaluable spatial-temporal data on different scales and in
different settings adding new layers of knowledge to urban studies, but the use of GPStechnology and deployment of GPS-devices still offers significant challenges for future research.
Besides, the enormous use of GPS tracking technologies hinges critically on the functioning of
the device.
Nowadays, the inside system of a portable low-cost GPS tracking device is being designed more
complex due to the requirements of precision and accuracy. Configurations of a GPS tracking
device when conducting field track are becoming more advanced and complicated. How well
does the current GPS device perform in tracking vehicle mobility, how much can we trust in the
accuracy information provided by manufactures? As argued by Shoval (2008), the tracking
device can function as an effective and reliable data collecting tool only if it does not affect the
nature, quality or authenticity of the data obtained.
Following this and the reviewed literatures, the assessment of the reliability of GPS tracking data
should be scrutinized. In this paper, we examine how well GPS tracking data matches the
travelled route for a bike, a car, and a bus for which the route, the speed, and the altitude are preset within the experiment. In the experiment, we vary the type of vehicle, speed, altitude,
sampling frequency, and filtering level.
Section 2 provides a review of GPS tracking data with a focus on works examining the reliability
of such data. Section 3 presents the experimental design and the data collect process. Section 4
gives the experimental results. Section 5 presents the discussions of the findings and conclusions.
2. Literature review
The global positioning system (GPS) has penetrated in transportation and is enlarging its
application thanks to the popularity of portable, low-cost GPS devices. There are three main parts
of applying GPS based device in transportation: collect travel data, analyze travel behavior and
evaluate performance of real field GPS data. Some seminal studies are enumerated in Table 1.
For the first aspect, several pilot studies have combined GPS technology with travel survey data
collection to improve the quantity and quality of travel data. Eby et al., (1997) used in-vehicle
GPS receivers to compare route choice behavior and perceptions of three different in-vehicle
navigation-assistance systems. Wagner (1997), Casas and Arce(1999), Draijer et al.(2000),
Doherty et al. (2001) respectively conducted a comprehensive data collection with GPS in
Lexington, Austin, Quebec City, and the Netherlands to test this method. Choi et al. (1998) and
Quiroga and Bullock (1999) concluded the in-vehicle GPS data collection system could be
applied as an efficient and economical tool for gathering comprehensive travel and traffic data.
Wolf (2000) studied in detail to use GPS data collection to completely replace rather than
supplement the traditional travel diary. Wolf et al. (2001) used GPS data loggers to collect travel
data in personal vehicles and demonstrated that it is feasible to derive trip purpose only from the
GPS data by using a spatially-accurate and comprehensive GIS. Based on the former work, Wolf
(2004) presented an overview of the various location collection technologies along with the
applications in travel behavior surveys, and on-developing technology trends. Leduc (2008)
conducted a snapshot of the development of traffic data collection methods and discussed the
potentials and bottlenecks related to emerged technologies as well as some short-term
perspectives. Considering that positioning technologies based on stand-alone GPS receivers are
vulnerable and have to be supported by additional information sources to obtain the desired
accuracy, integrity, availability, and continuity of service; Skog and Handel (2009) conducted an
in-depth survey of the information sources and information fusion technologies used in current
in-car navigation systems.
Table 1: Some empirical studies of GPS based data in transportation application
Research question of
GPS based data in
transportation
Literature
GPS-integrated
collection of travel
data
Choi et al.(1998), Casas and Arce(1999), Draijer et al.(2000), Dohertyet al. (2001), Ebyet al.,
(1997), Leduc (2008), Quiroga and Bullock(1999), Skog and Handel (2009), Wagner (1997),
Wolf (2000), Wolf et al. (2001), Wolf (2004)
Travel behavior
analysis based GPS
travel data
Askbrook and Starner (2003), Cooper et al.(2010), Feng and Timmermans (2013), Grengs et
al.(2008), Krumm and Horvitz (2006), Liao et al.(2004, 2005), Li et al. (2004, 2005), Oliver et
al.(2010), Patterson et al.(2003), Sun and Ban (2013), Troped et al.(2008), Schönfelder et al.
(2006), Huang and Levinson (2012), Wang et al. (2013), Wolf et al.(2003), Wolf et al.(2006);
Zheng et al. (2008), Zhang and Levinson (2008),
Evaluation
performance of GPS
travel data
Bhatti and Ochieng(2007), Chen and Li (2004), Dixon (2005), Du and Barth(2004, 2006), Enge
and Misra(1999), Farrel and barth (1999), Godha and Cannon, (2007), Hein(2000), Herrera et al.
(2006), Hounsell et al. (2008), Huang and Tan(2006); Lapucha et al.(2006), Modsching et
al.(2006), Misra et al.(1999), Marias et al.(2005), Obradovic et al.,(2006, 2007),Ochieng et al.
(2003), Quddus et al.(2006, 2007), Sanwal and Walrand (1995), Schönfelder and Antille (2002),
Schlingelhof et al. (2008), Tsakiri et al. (1998) , Yang and Farrell (2003), Zito et al. (1995), Yim
and Cayford (2001), Zhao et al. (2011)
For the second aspect, collecting individuals’ travel data lasting for a long period of time for
studying travel behavior has become possible by using GPS technology, especially those
commercial GPS devices in the current marketplace. The prime advantage of using GPS is that it
incorporates real-time spatial and temporal information throughout the entire trip (Wolf et al.,
2003; Grengs et al., 2008) based on which we are able to identify travel time and distance, origin
and destination pair with link-by-link trajectory, commute start and end times, and trip itineraries.
Askbrook and Starner (2003), Krumm and Horvitz (2006), Liao et al.(2006, 2007) and Patterson
et al.(2003) aimed to understand individual outdoor movements from GPS data. In which,
Askbrook and Starner (2003, Liao et al. (2006) extracted significant places of an individual from
GPS data-based activity, Krumm and Horvitz (2006) predicted a person’s movement. Patterson et
al. (2003) applied GPS tracks to classify a user’s mode of transportation as either “bus”, “foot”,
or “car”, and to predict his or her most likely route. Similarly, Liao et al. (2007) aimed to infer an
individual’s transportation routine given the person’s GPS data. Their system first detected a
user’s set of significant places, and then recognizes the activities like shopping and dining among
the significant places. Zheng et al. (2008) propose a graph-based post processing approach based
on supervised learning to infer people’s motion modes from their GPS logs. In their following
work on 2010, they proposed a more comprehensive approach combing a change point-based
segmentation method, an inference model and a graph-based post-processing algorithm to
automatically infer users’ transportation modes, including driving, walking, taking a bus and
riding a bike, from raw GPS logs.
Li et al. (2004) inspected the travel time variability in commute trips, the effects on departure
time and route choice. Li et al. (2005) investigated attributes whether to choose one or more
routes in the morning commute in attributes determining. Wolf et al. (2003) studied the underreporting trips in household travel surveys and Zhang and Levinson (2008) compared the impact
of information with other variables such as travel time, distance and aesthetics and then estimated
the value of information for travelers. Schönfelder et al. (2006) concluded that instrumented
vehicle GPS data can provide unique insights into the structure, size, and stability of human
activity spaces. Huang and Levinson (2012) analyzed the impacts of travelers’ interactions with
road network structure and clustering of services at the destinations on their retail destination
choice based on GPS travel data in the Twin Cities. They revealed that higher accessibility and
diversity of services around destinations are more attractive.
Wang et al. (2013) listed out studies related to non-work activity recognition based on GPS data
and examined the spatial patterns of non-work activities for 34 drivers in the Southeast Michigan
region. They found a strong dependence of non-work activity locations on commuting distances,
and an influence of commuting routes on non-work activities chained in all types of travel. The
results underline the importance of commuting routes in shaping the spatial configuration of nonwork activities. Sun and Ban (2013) proposed methods to classify vehicles using GPS data and
found that features related to the variations of accelerations and are the most effective in terms of
vehicle classification using GPS data. Recent research has attempted to combine GPS and
accelerometer data to recognize physical activities (Wolf et al., 2006; Troped et al., 2008; Cooper
et al., 2010; Oliver et al., 2010). Feng and Timmermans (2013) examined employing
accelerometer data in combination with GPS data in transportation mode identification and found
substantial contribution to successful imputation.
For the third aspect, one non-negligible aspect is that tracking technologies based on a single
GPS receiver are vulnerable and, thus, have to be supported by additional sensor information
(Skog and Handel, 2009). Farrel and barth (1999) gave the typical standard deviation of mode
errors in the ranging estimates of a single-frequency GPS receiver, working in standard precision
service mode. Ochieng et al. (2003) and Bhatti and Ochieng(2007) listed possible GPS failure
modes. Some seminal studies therefore conducted evaluation on the performance of the GPS
travel data in real field to check its reliability, validity and integrity.
Sanwal and Walrand (1995) addressed some of the key issues of a traffic monitoring system
based on probe vehicle reports (position, speeds, or travel times), and concluded that they
constitute a feasible source of traffic data. Zito et al. (1995) also investigated the use of GPS
devices as a source of data for traffic monitoring. Two tests were performed to evaluate the
accuracy of the GPS as a source of velocity and acceleration data. Yim and Cayford (2001) used
a vehicle equipped with differential GPS (DGPS), and managed to match its route for 93% of the
distance it traveled. Herrera et al. (2006) suggested that a 2–3% penetration of GPS-based cell
phones in the driver population is enough to provide accurate measurements of the velocity of the
traffic flow. Zhao et al. (2011) concluded that based on GPS spot speeds was sufficiently accurate
in comparison with space mean speeds, with a mean absolute difference of less than 6%.
Tsakiri et al. (1998) and Hounsell et al. (2008) respectively, discussed on robustness
enhancement of a bus fleet monitoring system and the use of GPS in bus priority control at traffic
lights. Schönfelder and Antille (2002) showed the selective inaccuracy of the available GPS data
requires additional data processing work by using a real field data. Followed Yang and Farrell
(2003), Schlingelhof et al. (2008) confirmed that development of intelligent transport system
(ITS) applications, such as advanced driver assistance systems (ADASs), traffic control,
automatic positioning of accidents, electronic fee collection and goods tracking requires not only
navigation systems with higher accuracy but also better reliability and integrity with auxiliary
information. For example, under normal conditions, the location and the trajectory of a vehicle
are restricted by the road network.
Map matching which use a digital map of the road network to impose constraints on the GPS
navigation therefore has become a popular solution (Quddus et al., 2006; Du and Barth, 2004,
2006; Obradovic et al., 2006, 2007). In map matching, by comparing the trajectory and position
information from the GPS receiver with the roads in the digital map, the most likely position of
the vehicle on the road is estimated (Skog and Handel, 2009). Quddus et al. (2007) conducted a
thorough survey of map matching algorithms and ideas on further research directions.
To give absolute figures on the accuracy of in-car navigation systems is difficult since the
performance of the systems depends not only on the characteristics of the sensors, GPS receiver,
vehicle model, and map information but on the trajectory dynamics and surrounding environment
as well (Skog and Handel, 2009). In urban environments, buildings may partly block satellite
signals, forcing the GPS receiver to work with a poor geometric constellation of satellites,
thereby reducing the accuracy of the position estimates (Marias et al., 2005; Huang and Tan,
2006; Modsching et al., 2006; Godha and Cannon, 2007). Marias et al., 2005 also found that
multipath propagation of the radio signal due to reflection in surrounding objects may lead to
decreased position accuracy without notification by the GPS receiver, thereby reducing the
integrity of the navigation solution.
Dixon (2005) found that many low-cost GPS receivers are able to use correction data from the
satellite-based augmentation systems (SBASs). Chen and Li (2004) concluded from their test
results that based on correction data from the wide area augmentation system (WAAS) and the
European geostationary navigation overlay service (EGNOS) system; demonstrate position
accuracy in the range of 1–2 m in the horizontal plane and 2–4 m in the vertical plane at a 95%
confidence interval. By using dual-frequency receivers and carrier-phase measurements
supported by various augmentation systems (SBASs, WAAS, EGNOS, MSAS), it is possible to
achieve real-time position accuracy on a decimeter level (Enge and Misra, 1999; Misra et al.,
1999; Hein, 2000; Dixon, 2005 Lapucha et al.,2006) However, the required receiver units are
currently far too costly for use in commercial in-car navigation systems, Dixon (2005) especially
discussed the performance and cost of single and dual-frequency GPS receivers and various
augmentation.
3. Experimental design and data collection
We want to examine how well GPS tracking data matches an actual route travelled. Vehicles are
in focus for this study and we therefore assume them being restricted by an underlying road
network. We consider the vehicles bike, car, and bus being the dominating means of private
transportations. In the experiment, the vehicles travel on pre-set routes of known geographical
position and altitude with speeds decided in advance. While they are travelling their mobility is
being tracked by a GPS device.
Figure 1: Interface of setting configurations for the GPS device BT-338(X)
We use the GPS device BT-338(X) which is a combination of a GPS receiver and a data logger1
According to the manufacturer, the device should provide a geographical positioning within an
error of 5 meters and a measurement error of velocity less than 0.4 km/h. The manufacturer
1
http://www.globalsat.com.tw/productspage.php?menu=2&gs_en_product_id=2&gs_en_product_cnt_id=20&img_id=414&product_cnt_folder=8
makes no claims about the precision in the measurement of altitude. Figure 1 illustrates the
interface in configuring the device with regard to some of the factors in the experiment. We
consider intensive sampling by the device with measurements every one and five seconds as well
as sampling every 30 seconds. Note that the latter implies that the vehicles will easily travel more
than 500 meters between recordings. Such setting implies a coarse assessment of the vehicle’s
mobility pattern. Hence, the levels of sampling frequency represent both dense and sparse data.
We set the data logging format to track position, time, date, speed, and altitude. The
WAAS/EGNOS/MSAS feature is enabled to acquire more precise position as suggested by the
manufacturer. We consider both enable and disable data logging when distance is less than the
selected radius 20 meters.
Table 3 illustrates the factors and corresponding levels in the experimental design. We are in
possession of 15 identical GPS devices with a unique identifying number. They devices are
randomly assigned to one of three groups of equal size for which the sampling interval is set to 1,
5, and 30 seconds respectively. In each group two randomly selected devices have the data
logging disabled if distance is less than the radius of 20 meters.
On the bike, all the 15 devices are carried by the rider in a backpack. Moreover, the devices are in
the backpack in the back seat of the car while the backpack is kept in the front seat of the bus.
The data collection of the bike and the car is undertaken in Borlänge in Sweden. The data
collection of the bus is undertaken along the bus line 151 between Borlänge and Falun in Sweden.
Table 3: Experimental design of collecting GPS tracking data
Samplin Interval
1s
Device No.
3
Distance Restriction
Distance
radius 0m
15km/h
20km/h
30km/h
Bicycle
40km/h
45km/h
50km/h
15km/h
20km/h
30km/h
40km/h
Car
45km/h
50km/h
60km/h
70km/h
Bus
80100km/h
29
37
5s
36
42
Distance
radius20m
4
14
Distance
radius 0m
39
30s
40
77
Distance
radius20m
9
32
Distance
radius 0m
74
24
72
Distance
radius20m
It was difficult to fix the velocity of the bus in advance as would be preferable. The velocity
varied along the scheduled route due to the traffic and the behavior of the drivers. For this reason,
only a segment of the route, where the velocity varied smoothly between 80 km/h and 100 km/h,
was used for GPS tracking. Meanwhile the bus trip was filmed. The bike followed a strict setting
of velocities ranging from 15-50 km/h in six levels. For the car a velocity of maximum 70 km/h
was considered. Travel diaries were used to note unexpected changes in route, velocity, and
emergent situation. The bike was ridden by the same rider and the driver of the car was the same
throughout the experiment.
Data for the bike was collected at noon in order to reduce the risk of deviation from the protocol
caused by other people on the route. Likewise, data collection for the car was undertaken between
3 and 4 in the afternoon to avoid peaks in the traffic. The data collection for the bus was
conducted after 6 in the afternoon thereby minimizing the variation in velocity due to people
waiting at bus stop. The data collection took part on a cloudy summer day with an air temperature
of about 22 degrees and almost no wind.
An accurate speedometer of the vehicles is essential for the experiment. To ensure this we first
considered the speedometer of the bike. The speedometer works by counting the wheel
revolutions per time unit adjusted by the circumference of the tire. Crucial for the accuracy is the
measurement of the circumference. The tires were inflated immediately prior to the experiment
and the circumference was measured by two different tape measurers. Thereafter we calibrated
the car speedometer by riding the bike and driving the car side by side and recording the speeds
simultaneously. We checked the relationship between the recordings from the bike speedometer
and the car speedometer by means of linear regression:
. The relationship is
strong with a correlation of 0.998. The speedometer of the car was adjusted accordingly in the
experiment.
The routes for the experiment were chosen having the need for maintaining a constant velocity in
mind. In the choice of routes, we tried to avoid places where the GPS signal was likely to be
disturbed. This means that the routes do not pass high buildings, strong magnetic fields or are in
valleys. As for the car, we also needed to consider the speed limits of the roads while a bike may
be ridden at any speed on a bike path.
Figure 2(a) depicts the route for the bike with arrows indicating the riding direction. The route is
about 2 kilometers and it is a paved bike path. The route was used consecutively for each velocity
at a time. For instance, at the velocity of 20 km/h the route took 6 minutes meaning that there
could be 360, 72, and 12 recordings per GPS device for the three levels of sampling frequency.
The variation in altitude of the route is only a few meters.
Figure 2(b) depicts the route for the car with arrows showing the directions. The route is
segmented by color representing the attained velocity. The route was travelled several times to
ensure sufficiently many recordings per cell in the experimental design. The range in altitude is
40 meters. Maintaining a constant velocity with a car in an ordinary traffic situation is of course
difficult. The circles in figure 2(b) represent segments identified in advance as impossible to
maintain the speed due to intersections and speed bumps. After the experiment recordings,
pertaining to segments where the intended velocity was not met according to the travel diary,
were removed. Figure 2(c) depicts the bus route. This route has the greatest variation in altitude
with a range of 37 meters.
Figure 2: (a) The bike route; (b) The car route; (c) The bus route
All the GPS devices were turned on before initiating the data collection. The reason was that
there is acquisition time for the device to start recording. The original GPS tracking data were
kept into DataLogger files. The files may be loaded from the device to a computer by using the
software GlobalSat Data Logger PC Utility. We retrieved the data directly after the experiment
was completed. The device number 4 was malfunctioning and did not record any data. The other
14 devices worked well and we obtained in total 25,901 recordings of the car, 9,224 recordings of
the bike, and 8,688 recordings of the buses.
As a final remark we note that there is a trade-off between sampling interval and battery lifespan
(Ryan et al., 2004). We checked whether the duration of the battery of the device differed for
various settings of the sampling interval. The check was conducted by randomly selecting 6 of
the GPS devices and letting 3 of them with intervals 1, 5, and 30 seconds and letting the other 3
of them with intervals 1, 5, and 30 seconds and data tracking within 20 meters distance radius
disabled. It turned out that the duration of the battery was unrelated to these two factors.
4. Experimental results
We begin by examining the positional reliability, followed by examining the reliability of
velocity and end with a check on the measurement of altitude obtained from the GPS device.
4.1 Geographical positioning
The geographical positions of the mobile object are necessary to identify its trajectory. In the
experiment the trajectory of the vehicles is known by the road network and its digital
representation. The location and the trajectory of a car are restricted by the road network (Skog
and Handel, 2009), as a statistics to assess the reliability of the geographical positioning obtained
from the GPS device we measure the concordance of the recordings and the road network. Ideally
the positional recordings should be on the underlying road network (national road data base of
Sweden, NVDB). Figure 3 shows by an example some of the positional recordings on the road
network. The red circles indicate the recordings that match the road network. The yellow circles
indicate recordings on the edge of the road network, by us regarded as matching the road network
well enough. The green squares indicate inaccurate recordings off the road network. In this
example, 8 of the 42 recordings failed in giving an accurate position of the car. The width of the
road is 14-20 meters meaning that an error of 5 meters is tolerated even if one considers that the
car was not driven in the middle of the road.
Figure 3: Example of positional recordings and the road network
A bike-path in NVDB is represented by a line, not a polygon, although its width is 3.5 meters
according to the department of motor vehicles in Sweden. In assessing the positional recordings
of the bike to the underlying road network we allowed for a tolerance distance of 5 meters.
Table 4 gives the proportion of positional recordings that match the road network. Considering
that the manufacturer of the GPS device claims that the error in positioning is at the most 5
meters, it is to be expected that almost all recordings should match the road network. This is
generally not the case. The positioning of the bus is accurate in 75-90% of the whole recordings.
The positioning of the car went fairly well with about 90% of the recordings being accurate. As
for the bike, the recordings frequently fail to identify its travel on the road network.
There is no clear pattern emerging from the factors considered in the experiment. Possibly the
longest sampling interval tends to lead to better positioning, the device generally gives higher
accuracy in positioning for the car but tends to have large variation on bike. However, we have
noted a serial correlation of the recordings implying that an inaccurate recording is likely to be
followed by another if the time interval is short.
Table 4: Proportion of positional recordings matching the road network
Vehicle
Bike
Car
Bus
Distance radius 0m
60.06%
94.97%
75.75%
Distance radius 20m
68.24%
91.15%
77.21%
Distance radius 0m
54.90%
87.33%
75.29%
Distance radius 20m
26.69%
93.27%
74.42%
Distance radius 0m
73.00%
92.15%
80.95%
Distance radius 20m
91.18%
92.86%
90.00%
Factors
1s
5s
30s
The surprising results for the bike prompted us to run a secondary experiment considering it is
easy to control and comparison experiment can be conducted on the car road as well. We
speculated that the positional recordings of the bike were interfered by the surrounding
environment. Figure 4 depicts the two routes travelled by the bike at a second occasion. One
route coincides with the route used in the original experiment while the second route is a part of
the car’s route.
In the first experiment, we had numerous inaccurate recordings in the three areas depicted in
Figure 4 by a white circle and two triangles. The circled area is nearby power lines to the north.
The areas indicated by triangles have trees with height of 8-10 meters. In the secondary
experiments all settings of the GPS devices were kept as in the first experiment, but the bike
travelled both routes at a speed of 20 km/h. Table 5 gives the proportion of accurate recordings
on the two routes. Although the proportion of accurate recordings on the original bike route is
higher in the second experiment, it is still rather low. Again most inaccurate recordings happened
at the three areas previously identified as complicated. The positional recordings on the car’s
route were substantially better. This exercise illustrates that the GPS device may generate
(infrequent) errors due to the interferences with the surroundings such as trees and built-ups in a
non-obvious way (Modsching et al., 2006).
Figure 4: Bike routes in the secondary experiment
Table 5: Proportion of positional recordings matching the road network for the bike in the secondary experiment
Vehicle
Original route
On the car’s route
Distance radius 0m
73.83%
89.22%
Distance radius 20m
58.79%
99.50%
Distance radius 0m
50.38%
90.06%
Distance radius 20m
69.29%
88.38%
Distance radius 0m
71.49%
98.78%
Distance radius 20m
80.13%
100%
Factors
1s
5s
30s
4.2 Estimating the velocity
It goes without saying that it is more difficult to estimate a changing velocity than a constant
velocity. Drivers (and riders) need to adjust their speed in line with the traffic but also at
intersections, roundabouts, tortuous locations (Jia et al., 2012) and traffic lights. This is also true
in conducting an experiment of this kind. We used the travel diary of the car and the bike to
delete recordings where the intended constant velocity was not possible to maintain. As for the
bus, the films were used for deleting recordings where the velocity was not constant.
Figure 5: Recorded velocity and actual velocity as measured by one GPS device for the car
Table 6: Statistics of recorded velocity for bike, car and bus
Figure 5 illustrates how the recorded velocity varies around the pre-set constant velocity. The
figure shows the recordings for one device in the car where the device was set to record the
velocity in intervals of 30 seconds. There is a tendency that the recorded velocity is generally
lower than the actual velocity. Recall that the manufacturer claimed that the error in velocity
should be within 0.4 km/h. Table 6 further shows the statistics for the recorded velocity as the
average, the standard deviation and the root mean square error. The velocity is underestimated by
about 5% and the standard deviation exceeds by far 0.4 km/h. The relative error in the recorded
velocity seems not to be related to the setting of the GPS device.
We have conducted analysis of variance (ANOVA) to formally test for the factors. The error
between the recorded and actual velocity was the response variable. The error increased with the
velocity. There was no significant difference for whether the distance restriction was on or off.
The sampling interval was unrelated to the error, except for the recordings of the bike. In this
case the longer sampling interval was associated with a (marginal) increase in the error.
We also check for a relationship between the error in velocity and the positional error as
discussed in section 4.1. We did so by labelling all positional recordings on the road network as
accurate and all those off the road network as inaccurate. Thereafter we repeated the ANOVA
including the factor Accurate in the model. It was strongly significant suggesting a greater
underestimation of the velocity if the positional recording was inaccurate.
4.3 Altitudes
The GPS device is presumably able to record the altitude of the vehicle as it travels. However, the
manufacturer is not specific about the precision in the recorded altitude. We expect the precision
of altitude to be poorer that the geographical position considering for instance the requirement for
connection to additional satellites for estimating altitude.
In order to check the precision in the recorded altitude, we first acquired the geo-information of
altitude in Borlänge, Sweden from NVDB. Thereafter we applied spatial join in Arc GIS 10.1 to
join the attribute table of the actual altitude layer to the attribute table of the recorded altitude
layer. Each position of the vehicle where a recorded altitude happened is related to the nearest
point in the actual altitude layer. The maximum distance between the position of the recording
and the actual altitude layer is 21 meters. This is an inconsequential approximation as the road
network covered in the experiment does not contain any steep up- and down-hills. Another
(trivial) approximation is the fact that the devices were carried by the rider in a backpack, in the
back seat of the car, and in the front seat of the bus. Hence, the altitude of the devices was 1-2
meters above the level of the road network.
The error in recorded altitude with respect to the actual altitude is large. Most of the time the
error was within the range of -50 meters and 50 meters, but frequently the error exceeded 100
meters. Considering for instance that the vehicle routes especially the bike paths travelled in the
experiment was essentially flat such a magnitude in error is enormous.
Moen et al. (1996) discussed the concepts of 2-D and 3-D fix and argued that a 3-D fix should
offer a greater precision in estimating the altitude. The GPS device used in the experiment
generates a 2-D fix. All the same, the results are not impressive.
5. Conclusion
This paper is focus on introducing a method of evaluating reliability of portable, low-cost GPS
device in tracking vehicle. The particular experiment is based on equipping GPS tracking device
BT-338(X) on vehicle being car, bike and bus and then track the geographical position, velocity
and altitude in the real road network.
Preprocessing and cleaning the data is necessary and auxiliary information is needed. The GPS
tracking data has fairly good performance in locating the actual positions of the vehicle in the
road network. The underlying road network examines that there are a certain proportion of the
data are off the road, the information provided by manufacturer is not precise in the practical
application. The circumstances that we conducted the experiments are quite simple, which do not
have high built-ups, obstacles, magnetic fields; the precision in positioning will probably become
worse when it goes to more complicated conditions. It verifies that the error of the GPS position
is dependent on the built-up and surrounding in the close vicinity (Modsching et al., 2006). This
can be detected and rectified by using map-matching algorithms, as proposed in the reviewed
literatures.
The tracked instantaneous velocities are quite accurate although there is a tendency of under
estimating the actual velocities. The error between recoded velocity and constant velocity is
monastically increasing as the velocity increases. The distance restriction configuration does not
have influence in the error. The error of bike is monastically increasing as the sampling interval
increases, while there is no obvious influence for the car and the bike. Considering the influence
of traffic flow, acceleration, variation of road network, the device provides acceptable accuracy
in velocity for transportation use. The devices do not immediately record the velocity to be 0
when the car and the bus stop. The reason might be due to that the device does not response
rapidly in the cooling down process of the car and bus. The altitudes in the tracking data are
inaccurate and not applicable considering the big variation of the errors. The approximation error
in spatial join should be responsible for this; however, some other potential reasons such as 3-D
fix (Griffin D., 2011), satellite constellations, integrate with other portable GPS devices and
circumstance changes need to be uncovered in the future.
Moreover, we always need to make a trade-off between the density of recordings and credibility
of data analysis. More intensive positional recordings do exhibit the movement pattern of a
mobile object in more detail. The travelled trajectories will match better to the real underlying
road network as well. However, we also obtain more aggressive data simultaneously which is
very time consuming to conduct data processing, data mining and analysis. We conclude that we
should set configurations of a GPS tracking device according to features of mobile object,
environment, road-network condition and data requirement. If we aim to investigate the mobility
from the geographical positions, a dense data set will fit better than a parse one. The time interval
can therefore be set as short as possible and no other restrictions, considering there will be
deletion in matching the road network. If the velocity change is the focus, the time interval can be
set to longer than 1 second, especially when a constant velocity can be easily kept for a while. If
we are interested in certain velocities, a feasible distance restriction or a velocity restriction can
be set to filter those velocities that are too low; but we need to consider the error of the GPS
device as well.
There is drawback of the GPS device due to the short effective lifespan (Ryan et al., 2004). The
experiments of data collection in this paper do not take longer than 2 hours each; the operation
time of the device is not a concern because it can have an operation time of 11 hours after fully
charged and in continuous mode. However, this can be a drawback if we want to acquire data in
time period much longer than 11 hours, it will induce to missing data and have significant cost of
data density and we should check in the future work. The GPS device BT-338(X) has SiRF Star
III as chip set and the frequency is L1, 1575.42 MHz, which is a specific type of GPS device,
devices with different electrical characteristics can be added for comparison.
References
1. Abbott, H., & Powell, D. (1999). Land-vehicle navigation using GPS. Proceedings of the IEEE, 87(1),
145-162.
2. Ashbrook D. & Starner T. (2003). Using GPS to learn significant locations and predict movement
across multiple users. Personal and Ubiquitous Computing, 7(5), 275-286.
3. Ashbrook D., & Starner T. (2002). Learning significant locations and predicting user movement with
GPS. In Wearable Computers, 2002. (ISWC 2002). Proceedings. Sixth International Symposium
on (pp. 101-108). IEEE.
4. Bhatti, U. I., & Ochieng, W. Y. (2007). Failure modes and models for integrated GPS/INS systems.
Journal of Navigation, 60(02), 327-348.
5. Brakatsoulas S., Pfoser D., Salas R., & Wenk C. (2005, August). On map-matching vehicle tracking
data. In Proceedings of the 31st international conference on Very large data bases (pp. 853-864).
VLDB Endowment.
6. Casas, J., & Arce, C. H. (1999, January). Trip reporting in household travel diaries: A comparison to
GPS-collected data. In 78th annual meeting of the Transportation Research Board, Washington,
DC (Vol. 428).
7. Chen, R., & Li, X. (2004). Virtual differential GPS based on SBAS signal. GPS solutions, 8(4), 238244.
8. Choi, K., Shin, C., & Park, I. (1998, August). An estimation of link travel time using gps and gis.
In Integrating the Transportation Business Using GIS. Proceedings of the 1998 Geographic
Information Systems for Transportation (GIS-T) Symposium.
9. Cooper, A. R., Page, A. S., Wheeler, B. W., Griew, P., Davis, L., Hillsdon, M., & Jago, R. (2010).
Mapping the walk to school using accelerometry combined with a global positioning system.
American journal of preventive medicine, 38(2), 178-183.
10. Doherty, S. T., Noël, N., Gosselin, M. L., Sirois, C., & Ueno, M. (2001). Moving beyond observed
outcomes: integrating global positioning systems and interactive computer-based travel behavior
surveys (No. E-C026).
11. Du, J., & Barth, M. (2006, June). Bayesian probabilistic vehicle lane matching for link-level invehicle navigation. In Intelligent Vehicles Symposium, 2006 IEEE , 522-527. IEEE.
12. Draijer, G., Kalfs, N., & Perdok, J. (2000). Global Positioning System as data collection method for
travel research. Transportation Research Record: Journal of the Transportation Research Board,
1719(1), 147-153.
13. Dixon, K. (2005). Satellite positioning systems: Efficiencies, performance and trends. European
Journal of Navigation, 3(1), 58-63.
14. Eby, D. W., & Kostyniuk, L. P. (1999). An on-the-road comparison of in-vehicle navigation
assistance systems. Human Factors: The Journal of the Human Factors and Ergonomics Society,
41(2), 295-311.
15. Enge, P., & Misra, P. (1999). Special issue on global positioning system. Proceedings of the IEEE,
87(1), 3-15.
16. Enge, P., Walter, T., Pullen, S., Kee, C., Chao, Y. C., & Tsai, Y. J. (1996). Wide area augmentation of
the global positioning system. Proceedings of the IEEE, 84(8), 1063-1088.
17. Farrell, J., & Barth, M. (1999). The global positioning system and inertial navigation (Vol. 61). New
York: McGraw-Hill.
18. Feng, T., & Timmermans, H. J. (2013). Transportation mode recognition using GPS and
accelerometer data. Transportation Research Part C: Emerging Technologies, 37, 118-130.
19. Grengs, J., Wang, X., and Kostyniuk, L. (2008). Using GPS Data to Understand Driving Behavior.
Journal of Urban Technology, 15(2):33–53., 1854:189–198.
20. Godha, S., & Cannon, M. E. (2007). GPS/MEMS INS integrated system for navigation in urban areas.
GPS Solutions, 11(3), 193-203.
21. Griffin, D. (2011, June). How does the global positioning system work. In GPS World Public
Forum (Vol. 26).
22. Hein, G. W. (2000). From GPS and GLONASS via EGNOS to Galileo–Positioning and Navigation in
the Third Millennium. GPS Solutions, 3(4), 39-47.
23. Herrera J. C., Work D. B., Herring R., Ban X. J., Jacobson Q., & Bayen A. M. (2010). Evaluation of
traffic data obtained via GPS-enabled mobile phones: The Mobile Century field experiment.
Transportation Research Part C: Emerging Technologies, 18(4), 568-583
24. Huang, A., & Levinson, D. (2012, February). Accessibility, network structure, and consumers’
destination choice: a GIS analysis of GPS travel data. InProceedings of the 91st Annual Meeting of the
Transportation Research Board. Transportation Research Board of the National Academies,
Washington, DC.
25. Hounsell, N. B., Shrestha, B. P., Head, J. R., Palmer, S., & Bowen, T. (2008). The way ahead for
London's bus priority at traffic signals. IET Intelligent Transport Systems, 2(3), 193-200.
26. Huang, J., & Tan, H. S. (2006). A low-order DGPS-based vehicle positioning system under urban
environment. Mechatronics, IEEE/ASME Transactions on, 11(5), 567-575.
27. Jia T., Jiang B., Carling K., Bolin M. & Ban Y. (2012). An empirical study on human mobility and its
agent-based modeling. Journal of Statistical Mechanics: Theory and Experiment, 2012(11), P11024.
28. Jia T., Carling K. & Håkansson J. (2013). Trips and their CO2 emissions to and from a shopping center.
Journal of Transport Geography, 33, 135-145.
29. Krumm, J., & Horvitz, E. (2006). Predestination: Inferring destinations from partial trajectories.
In UbiComp 2006: Ubiquitous Computing, 243-260. Springer Berlin Heidelberg.
30. Lapucha, D., Barker, R., & Zwaan, H. (2005). Wide area carrier phase positioning. European Journal
of Navigation, 3(1), 10-16.
31. Leduc, G. (2008). Road traffic data: Collection methods and applications. Working Papers on Energy,
Transport and Climate Change, 1, 55.
32. Li, H., Guensler, R., and Ogle, J. (2005). Analysis of morning commute route choice patterns using
global positioning system-based vehicle activity data. Transportation Research Record: Journal of the
Transportation Research Board, 1926:162–170.
33. Li, H., Guensler, R., Ogle, J., and Wang, J. (2004). Using global positioning system data to understand
day-to-day dynamics of morning commute behavior. Transportation Research Record: Journal of the
Transportation Research Board, 1895:78–84.
34. Liao, L., Patterson, D. J., Fox, D., & Kautz, H. (2006). Building personal maps from GPS data. Annals
of the New York Academy of Sciences, 1093(1), 249-265.
35. Liao, L., Patterson, D. J., Fox, D., & Kautz, H. (2007). Learning and inferring transportation routines.
Artificial Intelligence, 171(5), 311-331.
36. Misra, P., Burke, B. P., & Pratt, M. M. (1999). GPS performance in navigation.Proceedings of the
IEEE, 87(1), 65-85.
37. Modsching M., Kramer R., & ten Hagen K. (2006, March). Field trial on GPS Accuracy in a medium
size city: The influence of built-up. In 3rd Workshop on Positioning, Navigation and Communication ,
209-218
38. Murakami, E., & Wagner, D. P. (1999). Can using global positioning system (GPS) improve trip
reporting? Transportation research part c: emerging technologies, 7(2), 149-165.
39. Marais, J., Berbineau, M., & Heddebaut, M. (2005). Land mobile GNSS availability and multipath
evaluation tool. Vehicular Technology, IEEE Transactions on, 54(5), 1697-1704.
40. Ochieng, W. Y., Sauer, K., Walsh, D., Brodin, G., Griffin, S., & Denney, M. (2003). GPS integrity
and potential impact on aviation safety. The journal of navigation, 56(01), 51-65.
41. Obradovic, D., Lenz, H., & Schupfner, M. (2006). Fusion of map and sensor data in a modern car
navigation system. Journal of VLSI signal processing systems for signal, image and video technology,
45(1-2), 111-122.
42. Obradovic, D., Lenz, H., & Schupfner, M. (2007). Fusion of sensor data in Siemens car navigation
system. Vehicular Technology, IEEE Transactions on, 56(1), 43-50.
43. Oliver, M., Badland, H. M., Mavoa, S., Duncan, M. J., & Duncan, J. S. (2010). Combining GPS, GIS,
and accelerometry: methodological issues in the assessment of location and intensity of travel
behaviors. Journal of Physical Activity and Health, 7(1), 102-108.
44. Patterson, D. J., Liao, L., Fox, D., & Kautz, H. (2003, January). Inferring high-level behavior from
low-level sensors. In UbiComp 2003: Ubiquitous Computing(pp. 73-89). Springer Berlin Heidelberg.
45. Quiroga, C. A., & Bullock, D. (1999). Travel time information using global positioning system and
dynamic segmentation techniques. Transportation Research Record: Journal of the Transportation
Research Board, 1660(1), 48-57.
46. Quddus, M. A., Ochieng, W. Y., & Noland, R. B. (2006). Integrity of map-matching algorithms.
Transportation Research Part C: Emerging Technologies, 14(4), 283-302.
47. Quddus, M. A., Ochieng, W. Y., & Noland, R. B. (2007). Current map-matching algorithms for
transport applications: State-of-the art and future research directions. Transportation Research Part C:
Emerging Technologies, 15(5), 312-328.
48. Ryan P. G., Petersen S. L., Peters G., & Grémillet D. (2004). GPS tracking a marine predator: the
effects of precision, resolution and sampling rate on foraging tracks of African Penguins. Marine
Biology, 145(2), 215-223.
49. Schönfelder S. & Antille N. (2002). Exploring the Potentials of Automatically Collected GPS Data for
Travel Behaviour Analysis: A Swedish Data Source. ETH, Eidgenössische Technische Hochschule
Zürich, Institut für Verkehrsplanung, Transporttechnik, Strassen-und Eisenbahnbau IVT.
50. Schönfelder, S., Li, H., Guensler, R., & Ogle, J. (2006). Analysis of commute Atlanta instrumented
vehicle GPS data: Destination choice behavior and activity spaces. ETH, Eidgenössische Technische
Hochschule Zürich, IVT, Institut für Verkehrsplanung und Transportsysteme.
51. Skog, I., & Handel, P. (2009). In-car positioning and navigation technologies—A survey. Intelligent
Transportation Systems, IEEE Transactions, 10(1), 4-21.
52. Sanwal, K., Walrand, J., 1995. Vehicles as Probes. California PATH Working Paper UCB-ITS-PWP95-11, Institute of Transportation Studies, University of California, Berkeley, CA.
53. Shoval N. (2008). Tracking technologies and urban analysis. Cities, 25(1), 21-28.
54. Sun, Z., & Ban, X. J. (2013). Vehicle classification using GPS data. Transportation Research Part C:
Emerging Technologies, 37, 102-117.
55. Schlingelhof, M., Betaille, D., Bonnifait, P., & Demaseure, K. (2008). Advanced positioning
technologies for co-operative systems. Intelligent Transport Systems, IET, 2(2), 81-91.
56. Theiss A., Yen D. C., & Ku C. Y. (2005). Global Positioning Systems: an analysis of applications,
current development and future implementations. Computer Standards & Interfaces, 27(2), 89-100.
57. Tsakiri, M., Stewart, M., Forward, T., Sandison, D., & Walker, J. (1998). Urban fleet monitoring with
GPS and GLONASS. Journal of Navigation, 51(03), 382-393.
58. Troped, P. J., Oliveira, M. S., Matthews, C. E., Cromley, E. K., Melly, S. J., & Craig, B. A. (2008).
Prediction of activity mode with global positioning system and accelerometer data. Medicine and
science in sports and exercise, 40(5), 972-978.
59. Van der Spek S., Van Schaick J., De Bois P. & De Haan R. (2009). Sensing human activity: GPS
tracking. Sensors, 9(4), 3033-3055.
60. Wagner, D. P. (1997). Lexington area travel data collection test: GPS for personal travel
surveys. Final Report, Office of Highway Policy Information and Office of Technology Applications,
Federal Highway Administration, Battelle Transport Division, Columbus.
61. Wang, X., Grengs, J., & Kostyniuk, L. (2013). Visualizing Travel Patterns with a GPS Dataset: How
Commuting Routes Influence Non-Work Travel Behavior. Journal of Urban Technology, 20(3), 105125.
62. Wolf, J. L., Oliveira, M. G. S., Troped, P., Mathews, C. E., Cromley, E. K., & Melly, S. J. (2006).
Mode and activity identification using GPS and accelerometer data. Transportation Research Board
85th Annual Meeting (No. 06-2443).
63. Wolf, J. (2000). Using GPS data loggers to replace travel diaries in the collection of travel
data (Doctoral dissertation, Georgia Institute of Technology).
64. Wolf, J., Guensler, R., & Bachman, W. (2001). Elimination of the travel diary: Experiment to derive
trip purpose from global positioning system travel data. Transportation Research Record: Journal of
the Transportation Research Board, 1768(1), 125-134.
65. Wolf, J. (2004, August). Applications of new technologies in travel surveys. In7th International
Conference on Travel Survey Methods, Costa Rica.
66. Wolf, J., Oliveira, M., and Thompson, M. (2003). Impact of underreporting on mileage and travel time
estimates: Results from global positioning system-enhanced household travel survey. Transportation
Research Record: Journal of the Transportation Research Board , 1854(1), 189-198.
67. Yim, Y. B., & Cayford, R. (2001). Investigation of vehicles as probes using global positioning system
and cellular phone tracking: field operational test. California Partners for Advanced Transit and
Highways (PATH).
68. Zito R., D’este G. & Taylor M. A. P. (1995) Global positioning in the time domain: how useful a tool
for intelligent vehicle-highway systems? Transportation Research C 3(4), 193–209.
69. Zumberge J. F., Heflin M. B., Jefferson D. C., Watkins M. M. & Webb F. H. (1997). Precise point
positioning for the efficient and robust analysis of GPS data from large networks. Journal of
Geophysical Research: Solid Earth (1978–2012), 102(B3), 5005-5017.
70. Zhang, L. and Levinson, D. (2008). Determinants of route choice and the value of traveler information:
A field experiment. Transportation Research Record: Journal of the Transportation Research Board,
2086:81–92.
71. Zheng, Y., Li, Q., Chen, Y., Xie, X., & Ma, W. Y. (2008, September). Understanding mobility based
on GPS data. In Proceedings of the 10th international conference on Ubiquitous computing (pp. 312321). ACM.)
72. Zheng, Y., Chen, Y., Li, Q., Xie, X., & Ma, W. Y. (2010). Understanding transportation modes based
on GPS data for web applications. ACM Transactions on the Web (TWEB), 4(1), 1.)
73. Zhao, W., Goodchild, A. V., & McCormack, E. D. (2011). Evaluating the accuracy of spot speed data
from global positioning systems for estimating truck travel speed. Transportation Research Record:
Journal of the Transportation Research Board, 2246(1), 101-110.