Transportation Research Record 905 69

Abridgment

Traffic Data-Collection Systems: Current Problems and

Future Promise
RICHARD W. LYLES AND JOHN H. WYMAN

Evaluations were undertaken of off-the-shelf automatic traffic data-collection purposes and does not necessarily apply to special-
systems to examine their capabilities in two specific areas: data collection for purpose systems used in a research context.
speed compliance and vehicle classification by type. A prototype system de- In general, systems used for speed monitoring or
veloped by the United Kingdom's Transport and Road Research Labora- for vehicle classification have four basic compo-
tory (TRRL) was also briefly tested. System performances in the speed mode nents: sensing devices (sensors), which provide the
were not outstanding although several systems showed promise, whereas in the
essential indication, or signal, of a vehicle's
vehicle-classification mode, performance was not good and most systems expe-
rienced serious problems. Significant problems were also encountered in the
presence and movementi detectors, which receive the
use of pneumatic tubes as sensors. In the classification mode most systems signals from the sensors and amplify and/or inter-
also suffered from inadequate classification schemes (i.e., number and defini- pret themi and the recorder and processor, which re-
tion of categories). However, the TRRL system performed quite well; it uses ceive the signals from the detectors and calculate
a more sophisticated classification scheme (proposed as an alternative to cur· the speed or assign a category, record that informa-
rent schemes) and incorporates inputs from both presence and axle sensors. tion, and perform whatever manipulations of the
basic data are necessary to present them in final
Transportation engineers deal continually with large form (e.g., individual vehicle speeds and frequency
amounts of traffic data generated in a variety of counts) • Although these components essentially
ways and for numerous purposes. Because many states serve the functions indicated, most. systems do not
are experiencing financial problems, data-collection actually have separable components beyond the sens-
activities are being reviewed from several perspec- ing device. In addition, all systems are currently
tives: whether the data are really needed, what ac- limited to using only one type of sensing device in
curacy is required, and what the best ways are to a given installation.
collect them. Institution of the national maximum Most systems are capable of producing either
speed limit (NMSL) and the attention brought by the speed or classification (by type) data but not both
Federal Highway Administration (FHWA) to developing concurrently. However, it should be noted that when
the Highway Performance Monitoring System (HPMS) a system is operating in the classification mode, a
have also focused attention on data-collection tech- speed calculation is made internally as a prerequi-
niques. site to classification.
Although several states have investigated and/or Sensors that are conunonly used include inductance
invested in a different automatic data-collection loops, pneumatic tubes, coaxial cables, and tape-
system, most data collection in the areas noted has switches, although the last is typically used only
been done manually. Not only is this a labor-inten- in research situations. Inductive loops are typi-
sive operation, but there are other problems that cally imbedded in the pavement and are thus perma-
may be introduced, e.g., seasonal bias of data and nent and hence most costly, whereas the tubes and
unreliability of observers. In spite of the appar- cables are used on the road surface and have shorter
ent need for labor- and cost-saving systems, until lives. Although commonly used, easy to install, and
recently no organization had undertaken a comprehen- fairly inexpensive, the tubes are most prone to dam-
sive review of currently available systems or their age: The inherent high visibility of the installa-
performance. In light of this, FHWA contracted for tion led to one of the most prevalent problems in
an extensive evaluation of such systems with regard recent testing in Maine--purposeful damage by truck-
to how well they could satisfy data needs in two ers (i.e., locking their trailer brakes and skidding
specific areas: collecting speed-compliance data over the installation). Hence, life expectancy is
(with regard to the NMSL) and classifying ·vehicles unpredictable at best. It should also be noted that
by type (e.g., on the basis of axle spacing and air leaks are often hard to find and intermittent
overall length) • operation of systems is possible. The latter is a
Manufacturers' names will not be used with regard serious problem since total volumes or populations
to which system was best or worsti rather, the dis- of classification categories may be incorrectly es-
cussion will be limited to conunents about system timated from data that appear to be good.
capabilities in general. [The exception to this The visibility of cables is apparently consider-
will be in the discussion of sensing devices (indi- ably less than that of tubes (comparative tests
rectly) and experimental systems developed in the showed that the tubes were always seen and damaged
public domain.] However, findings with regard to by truckers, whereas no attempts were made on cable
specific systems may be found in final reports to sensor installations). Easy to install, the cables
FHWA (,!,ll . Systems that were evaluated were pro- are longer lasting than tubes and will resist pur-
duced by the following manufacturers: Leupold and poseful damage but will eventually succumb to snow-
Stevens, Inc. i Safetran Traffic Systems, Inc. i plows, studded tires, and sharp dragging objects
Streeter-Ameti Redland Automation, Inc. (Sarasota that breach their protective rubber coating. Varia-
Division) i and Golden River Corporation. A proto- tions of the cables (T-shaped cross sections imbed-
type unit from the United Kingdom's Transport and ded in pavement) have been developed and used by
Road Research Laboratory (TRRL) was also tested as TRRL with considerable long-term success.
part of the program but will not be discussed di- There are several other sensing devices that have
rectly. at least some potential for collection of vehicle-
classification and/or speed data, including the
GENERAL DESCRIPTION self-powered vehicle detector ( 3, 4) , the magnetic-
gradient vehicle detector (5,6), optical sensing
The discussion that follows is directed to available (7), audio signals (8), and electronic timers (9).
off-the-shelf systems that are marketed for general R;dar can also be used for speed data collection i1-
70 Transportation Research Record 905

though it is generally conceded that for large-scale Problems with Other System Components
collection programs it is cost-inefficient; it has
limitations in high-volume situations in being able The following are problems with other system compo-
to record the speeds of specific vehicles in queues nents:
and in introducing bias toward lower speeds given
the proliferation of citizen-band radios and radar 1. System-loop detector adjustments can be crit-
detectors. Although potential use of other such ical (and ace variable by site).
sensors is not barred, their use in the near future 2. There are numerous minor breakdowns.
in automatic systems seems unlikely. 3. Missed or misclassified vehicles are not ob-
viously due (in some instances) to sensor failures.
PROBLEMS WITH EXISTING SYSTEMS 4. Classification schemes ace typically simplis-
tic.
The existing data-collection systems were subjected
to a series of tests to evaluate their performance Positive Aspects of Systems
in field situations, i.e., how well they performed
in general and whether they did what they were pur- The existing systems are not without positive as-
ported to do. The systems were tested on separate pects. For example, it is clear that the technology
occasions in the speed mode and in the vehicle-clas- exists to process information from either axle or
s ification mode. presence sensors. In addition, some fairly sophis-
ticated differentiations among vehicles were incor-
Results in Speed Mode porated into the different systems' classification
schemes.
Results in the speed mode were as follows:
CURRENT AND PROPOSED VEHICLE-CLASSIFICATION SCHEMES
1. Primary problems were missed vehicles and in-
accurate speeds. As part of the larger question of what data are
2. Inaccurate speeds adversely affected speed- really needed, there is some debate over what con-
compliance percentages; there was a general high stitutes an adequate vehicle-classification (by
skew that was especially noticeable in the tails of type) scheme (10). For currently available systems,
the speed distribution (e.g., percentage over 55, the number of available categories varies from four,
65). based entirely on overall . length (raw data from
3. There was variation in accuracy from unit to loops), to eight, based on the number of axles and
unit (same manufacturer) or from site to site (same axle spacings (raw data from axle sensors). Users
unit). of vehicle-classification data, however, appear to
4. When test system was compared with base sys- desire more detail than is currently being pro-
tem(s) such as radar, fifth wheel, and/or optical vided. For example, FHWA has examined schemes with
timer, average speeds of samples of vehicles were from 7 to 32 categories and currently suggests 13 in
typically within l mph. the Highway Performance Monitoring System (HPMS)
program (11). At this point, then, there appears to
be a significant gap between what available systems
Results in Vehicle-Classi£ication Mode can deliver and what users desire.
The currently proposed schemes themselves also
The following results were found for the vehicle- have some problems since they do not necessarily re-
classification mode: flect logical differentiations among vehiCle types.
Some of the problems with currently di11cu111111d
schemes are summarized below (based on FHWA-supplied
1. Classification schemes were based only on
definitions and 1977 FHWA truck-weight study data) :
either overall length (loop based) or number of ax-
les and spacing (axle sensoc based, e.g., pneumatic 1. There is substantial overlap in classifica-
tubes) • tions of automobiles; with the trend in downsizing,
2. Sensitivity of loop detectors caused signifi- ourrcnt cutoff!! (e.g., !!Ubcompacl venius slctrn1arcl ur
cant errors. compact) are inappropriate--for wheelbase, < 100
3. Minimum length error: 5-8 percent of vehi- in; for overall length, < 180 in.
cles were not measured within ±5 ft. 2. There is overlap between vans, light pickups,
4. Maximum length error: 82 percent of vehicles and standard sedans or station wagons.
were not measured within ±5 ft (loop detector out
of tune) • 3. There is overlap between light trucks and
pickup trucks.
5. Minimum percentage of ax la-based classif ica-
4. No system is accurate on motorcycle differen-
tions: 15-20 percent of vehicles were misclassified.
6. Maximum percentage of axle-based class if ica- tiation.
tions: two-thirds of vehicles were misclassified. 5. Bus categories overlap with some trucks1
buses are seriously overestimated.

Sensor Problems In an effort to provide a somewhat more reason-

able departure point for further discussion of vehi-
The sensor problems included the following: cle classification, the scheme in Table l is pre-
sented. It is characterized by the following: (a)
it is based on information that can be obtained from
1. Axle-sensor-based units were limited to the axle sensors alone, (b) it recognizes some obvious
number of axles and axle spacings. problems with some differentiations and eliminates
2. Loop-based units were limited to overall them, and (c) it considers the frequency of encoun-
length only. tering certain types of vehicles (i.e., how impor-
3. Nature of axle sensors was nonpermanent. tant a specific category is). Three comments are
4. Undetected intermittent failure of axle sen- pertinent. First, the scheme is based on data from
sors (especially tubes) is possible. FHWA and assumes that the shortcomings of those data
5. Tubes are highly visible and often purpose- do not seriously compromise the characteristics of
fully damaged. different types of vehicles (e.g., some vehicle
Transportation Research Record 905 71

Table 1. Vehicle-classification proposal (14 categoriesl. enough to accommodate minor changes in the number
and parameters of categories.
Vehicle 6. All systems should have at least the option
Category Description Proposed Rule of printing out hourly data (inherently useful) so
that the time of potential breakdowns can be esti-
E-1 Passenger cars, light trucks, Axles= 2; wheelbase.;;; 10 ft mated.
vans 7. All systems should have straightforward and
E-2 Heavy-duty pickups, delivery Axles= 2; wheelbase> 10 ft
trucks, 2A6T's fully documented diagnostics, calibrations, and ad-
E-3 Cars and light trucks with one- Axles= 3 or 4; 1,2 spacing justments that can be made by the user.
or two-axle trailers .;;; 10 ft; 5.5 ft< 2,3 spacing 8. Systems that can simultaneously classify ve-
< 22 ft hicles by type and collect speed data should be in-
E-4 Three-axle single-unit trucks Axles= 3 and not E-3 vestigated.
E-5 Trucks and semitrailers (2S2) Axles= 4 and not E-3; 3 ft.;;; 3,4
spacing .;;; 10 ft
E-6 Four-axle single-unit trucks Axles= 4 and not E-3; 3 ft .;;; 2,3 The Automatic Vehicle Classification System
spacing .;;; 5 ft (AVCS) developed by TRRL and tested in Maine (12-16)
E-7 Other four-axle combinations Axles= 4 and not E-3, E-5, meets many of the above requirements, and the test
and E-6 results clearly demonstrated that reasonably sophis-
E-8 Trucks and semitrailers (3S2) Axles = 5; 2 ft .;;; 4,5 spacing
.;;; JO ft ticated traffic data-collection systems can be de-
E-9 Other five-axle combinations Axles= 5 and not E-8; 3 ft.;;; 2,3 veloped and used successfully in the field and that
spacing .;;; 5 ft the resulting data will have applicability in sever-
E-10 Trucks and semitrailers plus Axles = 5 and not E-8 or E-9 al areas of concern to engineers, planners, and pol-
full trailers (2Sl-2)
E-11 Trucks and semitrailers plus Axles = 6 and 5 ,6 spacing > 7 ft
icymakers.
full trailers (3Sl-2) There are, however, several questions that remain
E-12 Trucks and semitrailers (3S3) Axles = 6 and not E-11; 4,5 to be addressed by FHWA, the states, and other po-
spacing .;;; 6 ft tential users of such systems and/or the resultant
E-13 Other six-axle combinations Axles= 6 and not E-11 or E-12 data:
E-14 Other seven-or-more-axle Axles = 7 or more
combinations
1. Is the demand for such sophisticated data so
Note: An optional category would be for 281 truck and semitrailer combination. widespread as to warrant the development of systems
capable of delivering them?
2. Can sufficient consensus be achieved among
the potential users on the form of required data so
types are underrepresented although it is assumed that basic parameters required for classification
that recorded wheelbases, etc., would not differ and minimum or maximum system capabilities can be
from a scientifically drawn sample of the type) • defined?
Second, the scheme does not attempt certain differ- 3. Will the states and other local and nongov-
entiations that could be made if overall length data ernment users pursue purchase of new systems if such
were also considered (e.g., buses could probably be
data are not required and system acquisition is not
better differentiated). Third and last, the pro-
posed scheme is presented only as a point of depar- supported by the federal government?
ture for discussion by data users and others and not
The last point to be made concerns the capabilities
as the ultimate scheme.
of systems that use more or less permanent sites
(i.e., AVCS or AVCS-type system) versus those that
CONCLUDING REMARKS AND RESEARCH SUGGESTIONS might be truly portable (i.e., that use temporary
road-surface sensors). Current sensor technology,
Speed calculation and vehicle classification by and even that only available in prototype form,
length, wheelbase, axle spacings, and number of ax- basically constrains sophisticated equipment to us-
les are conceptually straightforward: e.g., the im- ing permanent sensor arrays that require a large
plicit decision rules for given categories are initial commitment of time and resources to imple-
easily stated in terms of overall length and/or the ment any comprehensive data-collection program: more
information from axle sensors. While the currently primitive systems can deliver lower-quality data
marketed systems tested typically used only a limit- much more cheaply. In this regard, it is not clear
ed number of categories, the more complex classifi- whether a truly portable system (including temporary
cation schemes do not necessarily imply the develop-
sensors) can provide the same quality of data as an
ment of new technology but rather a refinement of
AVCS-type system in a permanent installation. It is
that which exists. There are, however, several
apparent that large quantities of good data can be
areas that would benefit from additional work and/or
collectedi are they worth the cost of acquisition?
attention. These are outlined below:

1. A permanent, or at least longer-lived, axle ACKNOWLEDGMENT

sensor should be developed. At this point, it ap-
pears that some derivation of the coaxial cable Considerably more detail on the work reported here
would be the most likely candidate. may be found in the final report submitted to FHWA.
2. Prototype systems should be developed to ex- We are indebted to Robin Moore and Brian Stoneman of
amine the problems, if any, in interfacing coaxial TRRL for their assistance in the setting up and
cables with existing systems. testing of the AVCS in Maine. The work was funded
3. A single classification scheme should be by FHWA and undertaken at the Maine Facility Labora-
developed for use by the states (for their own pur- tory by personnel from the Maine Department of
poses) and FHWA. Such a scheme should recognize, Transportation and the Social Science Research In-
for example, trends in vehicle sizes, especially in stitute of the University of Maine and by us. The
the passenger-car categories. contributions of the facility staff and others not-
4. Systems need to be developed that can process withstanding, the responsibility for the material
inputs from a variety of sensing devices (e.g., from presented here and in the final reports rests with
only coaxial cables or from a cable-and-loop combi- us and does not necessarily represent the opinions
nation). of FHWA, Maine Department of Transportation, ~r any
5. System internal logic needs to be versatile other institution.
72 Transportation Research Record 905

REFERENCES for Vehicle Counting and Determining Vehicle

Miles of Travel--A Review of the Current State-
1. R.W. Lyles and J.H. Wyman. Evaluation of Speed of-the-Art on Vehicle Classification Programs.
Monitoring Systems. FHWA, Final Rept. FHWA/ NCHRP, draft report, Oct. 1979.
PL/80/006, July 1980. 11. Highway Performance Monitoring System: Case
2. R.W. Lyles and J.H. Wyman. Evaluation of Vehi- Study Procedural Manual--Vehicle Classifica-
cle Classification Equipment. FHWA, Final tion. Program Management Division, FHWA, Jan.
Rept., July 1982. 1980.
3. J.F. Scarzello and G.W. Usher, Jr. Self-Power- 12. R.C. Moore. Road Sensors for Traffic Data Col-
ed Vehicle Detector: Magnetic Sensor Develop- lection. Sensors in Highway and Civil Engi-
ment. FHWA, Rept. FHWA-RD-76-147, June 1976. neering. Institute of Civil Engineers, London,
4. J.H. Wyman and R.W. Lyles. Evaluation of the England, Paper 7, 1981.
Self-Powered Vehicle Detector. FHWA, June 1981. 13. R.C. Moore, P. Davies, and P.R. Salter. An
5. M.K. Mills. Magnetic Gradient Vehicle Detec- Automatic Method to Count and Classify Road Ve-
tor. IEEE Transactions on Vehicular Technol- hicles. Presented at International Conference
ogy, Vol. VT-23, No. 3, Aug. 1974. of Road Traffic Signalling, Institute of Elec-
6. M. Singleton and J.E. Ward. A Comparative trical Engineers, London, England, March-April
Study of Various Types of Vehicle Detectors. 1982.
U.S. Department of Transportation, Rept. DT- 14. Automatic Method to Classify, Count, and Record
TSC-OST-77-9, Sept. 1977. the Axle Weight of Road Vehicles. U.K. Trans-
7. T. Takagi. Optical Sensing and Size Discrimi- port and Road Research Laboratory, Crowthorne,
nation of Moving Vehicles Using Photocell Array Berkshire, England, Leaflet 639, 1979.
and Threshold Devices. IEEE Transactions on 15. Road Sensor for Vehicle Data Collection: The
Instrumentation and Measurement, Vol. 25, No. Axle Detector. U.K. Transport and Road Re-
1, March 1976. search Laboratory, Crowthorne, Berkshire, Eng-
8. K. Jakus and D.S. Coe. Speed Measurement land, Leaflet 905, 1979.
Through Analysis of the Doppler Effect in Ve- 16. Automatic Vehicle Count and Classification:
hicular Noise. IEEE Transactions on Vehicular Examples of Data Output. U.K. Transport and
Technology, Vol. VT-24, No. 3, Aug. 1975. Road Research Laboratory, Crowthorne, Berk-
9. M.R. Parker, Jr. An Evaluation of the Follow- shire, England, Leaflet 884, 1979.
ing Too Closely System. Traffic Engineering,
Vol. 46, No. 7, July 1976. Publication of this paper sponsored by Committee on Traffic Flow Theory and
10. John Hamburg and Associates. Improved Methods Characteristics.

Analysis of TRANSYT Platoon-Dispersion Algorithm

NAGUI M. ROUPHAIL

The development of an analytical solution to the recursive platoon-dispersion tion in TRANSYT-type models is platoon-dispersion
formula used in TRANSYT models of traffic flow is presented. Flow rates in behavior. Simply stated, as a queue of vehicles
the predicted platoon measured at the kth interval of the jth simulated cycle leaves the stopline on the green indication, its
are expressed in terms of demand and capacity rates at the source intersection shape is altered along the downstream link in a man-
in addition to signal-control and travel-time parameters. It was found that the ner reflective of the desire of individual drivers
TRANSYT recursive formula implicitly contains a cycle factor that results in
an underestimation of the toal flow rate simulated. An estimate of that error
to maintain comfortable time headways. Thus, al-
has been formulated, which can be applied as a constraint on the required simu- though the flow rate at the stopline is equivalent
lation time in TRANSYT. The analytical solution also provided insight into to the saturation rate in the presence of a queue
the determination of critical intersection spacings below which signal coordina- and to the demand rate thereafter (assuming un-
tion becomes feasible. dersaturated operation) , the flow patterns measured
at an observation point t seconds downstream of the
The proliferation of digital computer model applica- stopline would be considerably different.
tions in the areas of traffic flow and control in Mathematically, platoon-dispersion behavior is
the past decade has led to the successful develop- expressed by the following recursive relationship:
ment of several widely used traffic signal opera-
tions models, such as Network Simulation Model IN(k + t) = F x OUT(k) + (! - F) x IN(k + t - 1) (1)
(NETSIM), Signal Operations Analysis Package (SOAP),
Traffic Network Study Tool (TRANSYT) , and Traffic where
Signal Optimization Program (SIGOP) <l-il.
TRANSYT, a program for traffic signal timing and IN(k + t) flow rate in kth time interval of
coordination initially developed in the United King- predicted platoon at observation
dom by Robertson (5), has been successfully applied pointi
at many intersections in Europe and the United OUT(k) flow rate in kth time interval of
States. The TRANSYT-7F version, for example, has original platoon at stoplinei
recently been used in the National Signal Timing t e times average platoon travel time
Optimization Project (6), which encompassed 11 from stopline to observation point
cities and approximately 500 signalized intersec- le is an empirical travel-time fac-
tions in the United States. tor expressed as ratio between travel
The fundamental principle of traffic representa- time of leading vehicle in platoon