10 0007 .pdf



Nom original: 10-0007.pdfTitre: Microsoft Word - 10-0007_Sensys_2009_11_13.docxAuteur: cmday

Ce document au format PDF 1.5 a été généré par PScript5.dll Version 5.2.2 / Acrobat Distiller 9.0.0 (Windows), et a été envoyé sur fichier-pdf.fr le 26/05/2011 à 23:00, depuis l'adresse IP 82.67.x.x. La présente page de téléchargement du fichier a été vue 1589 fois.
Taille du document: 443 Ko (33 pages).
Confidentialité: fichier public

Aperçu du document


Day, Premachandra, Brennan, Sturdevant, and Bullock

Operational Evaluation of Wireless
Magnetometer Vehicle Detectors at a
Signalized Intersection
by
Christopher M. Day
Purdue University
Hiromal Premachandra
Purdue University
Thomas M. Brennan
Purdue University
James R. Sturdevant
Indiana Department of Transportation
Corresponding author:
Darcy M. Bullock
Purdue University
550 Stadium Mall Dr
West Lafayette, IN 47906
Phone (765) 496-7314
Fax (765) 496-7996
darcy@purdue.edu
November 13, 2009
TRB Paper 10-0007
Word Count: 4055 words + 12 x 250 words/Figure-Table = 4055 + 3000 = 7055

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 1 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

ABSTRACT
Compact wireless magnetometers are attractive vehicle detection for signalized intersections
because their installation requires minimal pavement cutting, and the detectors are less likely
than saw-cut inductive loops to malfunction because of pavement failure. A study evaluating the
performance of wireless magnetometers for signalized intersection operation was performed at
an instrumented intersection. A testbed was constructed with co-located inductive loop and
wireless magnetometer detection zones. A five-day analysis period was conducted for each of
two left turn pockets at an actuated coordinated signalized intersection. Discrepancies between
the detection and non-detection states were quantified using high resolution traffic event log
data, and 240 hours of data collection groundtruthed by visual inspection of video recordings of
the detection zones. Detector state change behavior was also characterized. Wireless
magnetometers detectors were found to perform similarly to loops in terms of missed calls, and a
slightly higher tendency to generate false detection calls. Detection state changes in the wireless
magnetometers had typical (85th percentile) reporting latencies of 0.2 seconds or less for
activation and 0.5 seconds or less for state termination. The paper concludes by recommending
8’ spacing of the sensors adjacent to the stop bar to minimize missed calls.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 2 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

INTRODUCTION
Inductive loop detectors (ILDs) are the most commonly used type of vehicle detection because of
their reliability in reporting vehicle presence. However, loop detector installation can be
expensive because of the physical connection required to connect the loops back to the traffic
cabinet. Such connections are sometimes infeasible at locations such as bridges and ramps.
Finally, saw-cut inductive loops are particularly sensitive to moisture and wire breaks associated
with pavement failure. Compact wireless magnetometers are a promising alternative to loops
because they require substantially less pavement cutting and require no physical connection to a
monitoring device.
A prototype two-axis magnetometer with wireless capability was developed at the University of
California Berkeley in 2003 (1). That technology was subsequently commercialized by Sensys
Networks, Inc. A number of detector installations had been established at various locations
worldwide as of 2008 (2). Several studies have been published with initial evaluations of the
technology (1,2,3,4) primarily for vehicle counting. In 2008, when the present study was
initiated, there were no rigorous studies reporting on the performance of these detectors for stop
bar detection. This paper reports on an evaluation of this wireless magnetometer detection
technology for stop bar detection at signalized intersections.

TESTBED DEVELOPMENT
The testbed for this study is located at the intersection of State Road 37 and Pleasant Street in
Noblesville, Indiana. Figure 1 shows a plan of the intersection with the relevant detector
configuration, including placement of loop detectors, wireless magnetometers, and the wireless
radio frequency (RF) signal path back to the cabinet. RF Repeaters I and II were installed in
March 2008 and RF Repeater III was installed in October 2008. The magnetometers were used
to operate the left turn phases at the intersection during the study period.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 3 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

Setback Detectors
RF Repeater II
(March 2008)

N

Inductive Loop
Wireless Magnetometer

RF Repeater III
(October 2008)

RF Link

Pleasant St.
(30 mph)

SBLT Detector Region
INDOT Cabinet
Purdue Cabinet

NBLT Detector Region

SR 37
(55 mph)

RF link to field
RF Repeater I (March 2008)
RF link
Access Point
Cable to cabinet
12 ft lane widths

Figure 1. Plan of testbed detector configuration (not to scale).

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 4 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

Figure 2 shows detailed plans of the left turn detection zones. Initially, one magnetometer was
installed in the center of each ILD (Figure 2a, M1, M2, M3, M4). A preliminary comparison of
detector and ILD response determined that the original configuration identified a large number of
vehicle passing over the magnetometers that were not detected. The cause of these missed
detections was traced to the wide spacing (Figure 2a, M1, M2, M3, M4) between sensors in the
magnetometer detection area, causing many stopped vehicles to be missed. On March 25, 2009,
three additional magnetometers were added to the left turn lanes (Figure 2b, M0, M1B, M2B).
M1A and M2A were then deactivated. The effect of these changes was to extend the detection
zone just beyond the stop bar, and to reduce the distance between the magnetometers near the
stop bar.
The ILD and magnetometer detection zones were co-located to allow direct comparison of
detector response to the same vehicles. Any interaction between the two detector types was
expected to be negligible, owing to the small size of the magnetometer relative to the large area
of the loops, and the negligible impact of the ILD on the magnetic field in the area of the
magnetometers. For example, a magnetic compass is only deflected if within approximately 1
inch of the ILD wire. At the test bed, the loop wire and the magnetometers were separated by at
least 1 ft. Since the field strength decreases by an inverse square relationship to distance, the ILD
magnetic field would be reduced by a factor of 144 for the magnetometers closest to the wire (1
ft).

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 5 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

L4
L4
3
6

6
M4

M4

15

9

15

L3

Loop Spacing

M3

9

15

L2
6

6
M3

Loop Spacing

6

Magnetometer Spacing

L3

9
17

L2
6
2

M2A

M2

Magnetometer Spacing

9

M2B

9

15

9

11

L1
L1
6

6
M1

3

M1B

2

M1A

3

8

3
Stop Bar

Stop Bar
M0

(a) Original NBLT installation.

(b) NBLT installation modified on March 25,
2009 by retiring sensors M1A and M2A and
adding sensors M0, M1B, and M2B.

Figure 2. Testbed geometry and detector placement for left turn lanes.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 6 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

DETECTOR EVALUATION METHODOLOGY
A considerable body of research has been published in recent years evaluating the performance
of video detection systems (5,6,7,8,9,10,11,12,13,14). As a result, there is now a large
contemporary body of work on detector evaluation forming a conceptual basis on which other
vehicle detection systems can be evaluated. This paper extends those concepts to the specific
application of evaluating wireless magnetometers.

Quantifying Discrepancies
One important concept is that of tracking discrepancies between different detector states. This
was introduced by Rhodes et al. (9, 10) in the evaluation of video detection systems using a
methodology where the video detector state was compared with the ILD state. This paper uses
similar concepts for comparing the detector states of the magnetometers and ILDs.
Figure 3 shows a typical plot of the magnetometer and ILD state during vehicle passage through
the detection zone. Each detector is considered to occupy either an on (“1”) or off (“0”) state.
Event data was collected using a high resolution (0.1 seconds) traffic signal data logger. Using
the notation LxiMyi, where Lxi represents the ILD state and Myi represents the magnetometer
state at time interval i, four possible combinations of detector states can be defined:


L0M0: ILD Off, Magnetometer Off. Both systems report that no vehicle is present.



L1M1: ILD On, Magnetometer On. Both systems report that a vehicle is present.



L1M0: ILD On, Magnetometer Off. ILDs report the presence of a vehicle while
magnetometers do not. This represents a potential false call by the ILDs, or a potential
missed call by the magnetometers.



L0M1: ILD Off, Magnetometer On. Magnetometers report the presence of a vehicle while
ILDs do not. This represents a potential false call by the magnetometers, or a potential
missed call by the ILDs.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 7 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

L0M0

L1M0

L1M1

L0M1

L0M0

On

Magnetometer State
Off
On

Loop Detector State
Off

Time

Figure 3. Illustration of loop (L) and Magnetometer (M) state types.

The L1M0 and L0M1 states indicate discrepancies between the magnetometers and the ILDs.
Figure 4a shows the duration of L0M1 discrepancies plotted over a 24 hour period for the
northbound left turn detection zone, while Figure 4b shows this plot for L1M0 type
discrepancies. A pan-tilt-zoom (PTZ) camera was used to establish a video record. The
correlation of detector discrepancies with the presence and/or absence of vehicles in the
detection zone was determined by manual inspection of the video during the discrepancy.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 8 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

90

Duration of Discrepancy (sec)

80
70

Discrepancy time during
Red
Yellow
Green

60
50
40
30
20

3 sec

10
0
0:00

2:00

4:00

6:00

8:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00 0:00
Time of Day

(a) L0M1 discrepancies, Northbound left turn, May 21, 2009.
90

Duration of Discrepancy (sec)

80
70

Discrepancy time during
Red
Yellow
Green

60
50
40
30
20

3 sec

10
0
0:00

2:00

4:00

6:00

8:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00 0:00
Time of Day

(b) L1M0 discrepancies, Southbound left turn, May 21, 2009.
Figure 4. Example 24-hour plots of discrepancy duration by type.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 9 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

Quantifying Latency
Figure 4 exhibits a large number of discrepancies with rather short durations. These are largely
due to slight variation in detection zone locations and stochastic variation in sensor response. A
range of sensor latency was proposed in 2007 specification by INDOT (15). Middleton et al. (11)
proposed using the statistical distribution of latency as a performance metric. For this study,
latency was estimated using the following formula:

  t EVENT,M  t EVENT, L 

xM  x L
,
v

Equation 1

where tEVENT,M represents the event time (activation or termination) of the magnetometer and
tEVENT,L represents the event time of the loop. The units of these terms are in seconds. The third
term is a correction for spatial difference in the detection zones: xM and xL are the positions of the
edge of the magnetometer (ft) and ILD zones, and v is vehicle speed (ft/sec). As shown in Figure
2b, there is a 3 ft spatial offset between the beginning and end of the magnetometer and the ILD
detection zones. Assuming a vehicle speed of 15 mph, a correction of –0.14 seconds is included
in the latency calculation.
To filter out lag due to detection zone boundaries as well as manage the amount of manual effort
to check discrepancies, a 3-second threshold was used to determine which discrepancies to
visually groundtruth. This 3-second threshold is shown by the dashed line in Figure 4a and
Figure 4b. Discrepancies shorter than 3-seconds were aggregated to build distributions of
detector call activation and termination latency. Discrepancies longer than 3 seconds were
subject to visual analysis. Detector call activation and termination events were identified from
characteristic sequences of the detector state. Figure 5 shows the six characteristic sequences for
identifying activation and termination latencies shown as l in Figure 5.


Activation, negative latency (Figure 5a): Negative activation latency was identified by
sequence {L0M0, L0M1, L1M1}. Magnetometers turned on before the ILDs.



Activation, zero latency (Figure 5b): Sequence {L0M0, L1M1} indicated when the
magnetometers and ILDs placed a call simultaneously.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 10 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock



Activation, positive latency (Figure 5c): Positive actuation latency was identified by
sequence {L0M0, L1M0, L1M1}. ILDs turned on before the magnetometers.



Termination, negative latency (Figure 5d): Negative activation latency was identified by
sequence {L1M1, L1M0, L0M0}. Magnetometers turned off before the ILDs.



Termination, zero latency (Figure 5e): Sequence {L1M1, L0M0} indicated when the
magnetometers and ILDs terminated a call simultaneously.



Termination, positive latency (Figure 5f): Positive activation latency was identified by
sequence {L1M1, L0M1, L0M0}. ILDs turned off before the magnetometers.

The distribution of activation and latency times is explained in detail in the next section.

l

l
On

On

Magnetometer State

Magnetometer State
Off
On

L0M0

L0M1

Off

L1M1

Loop Detector State

On

L1M1

L1M0

L0M0

Loop Detector State
Off

Off

Time

Time

a) Activation, negative latency

d) Termination, negative latency

On

On

Magnetometer State

Magnetometer State
Off
On

L0M0

Off

L1M1

Loop Detector State

On

L0M0

L1M1

Loop Detector State
Off

Off

Time

Time

b) Activation, zero latency

e) Termination, zero latency

l

l

On

On

Magnetometer State

Magnetometer State
Off
On

L0M0

L1M0

Loop Detector State

Off

L1M1

On

L1M1

L0M1

L0M0

Loop Detector State
Off

Off

Time

Time

c) Activation, positive latency

f) Termination, positive latency

Figure 5. Illustration of activation and termination latencies.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 11 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

TARGET PERFORMANCE THRESHOLD
Table 1 presents the target performance thresholds for vehicle detectors proposed in the 2007
INDOT specification (15). Requirements for “low” and “standard” performance are presented in
terms of the percentile values in the distribution of latency during red and yellow, and during
green intervals. For example, for standard performance the specification requires the 85th
percentile value of activation latency to be less than or equal to 1 sec during red or yellow, and
0.1 sec during green. In addition, the table indicates the maximum number of missed calls and
false calls that are acceptable during one-hour and 24-hour periods. False calls are not broken
down by phase state, and the acceptance criteria indicate that no more than 20 false calls per 1hour or 24-hour period are acceptable.
Table 1. Target performance thresholds according to INDOT specifications as defined by ITM No.
934-08P (15).

Test Parameter
Activation Response Time, Typical (Ra85%)
Activation Response Time, Maximum (Ra100%)
Termination Response Time, Typical (Rt85%)
Termination Response Time, Maximum
(Rt100%)
Number of Missed Calls, 24 hours
Number of Missed Calls, 1 hour
Number of False Calls, 24 hours
Number of False Calls, 1 hour

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Low Performance
Red/Yellow
Green
Interval
Interval
≤ 2.0 s
≤ 1.0 s
≤ 10.0 s
≤ 5.0 s
≤ 2.0 s
≤ 1.0 s
≤ 10.0 s
0
0

Page 12 of 30

≤ 5.0 s
≤ 10
≤ 10

≤ 20
≤ 20

Standard Performance
Red/Yellow
Green
Interval
Interval
≤ 1.0 s
≤ 0.1 s
≤ 5.0 s
≤ 1.0 s
≤ 1.0 s
≤ 0.1 s
≤ 5.0 s
0
0

≤ 1.0 s
≤ 10
≤ 10

≤ 20
≤ 20

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

MAGNETOMETER PERFORMANCE FINDINGS
Data were collected at the site between May 21 and June 1, 2009. No traffic incidents or other
adverse conditions were observed during the data collection period. Because the PTZ camera
could be aimed at only one detection zone at a time, 5 days of data (120 hours) were analyzed for
each of the northbound and southbound left turn approaches.

False Calls and Missed Calls
All discrepancies between the loop detectors and wireless magnetometer larger than 3 seconds
were classified as either false or missed calls for a particular device. Each discrepancy was
classified as shown in Table 2.
Three categories of false calls for each detector technology were determined from video
observation:


The active detector could become “stuck on,” holding the detect state for several seconds
after vehicles have left the detection zone. The operational impact was the same as that of
a false call. This was only observed for the magnetometers. This is discussed in more
detail later.



The active detector could present a false call due to adjacent lane activity, as confirmed
by lane incursions or the presence of wide vehicles. This occurred more often for the
ILDs than for the magnetometers. This is not unexpected because the ILDs are physically
more sensitive to lane incursions.



The active detector could present a false call without a verifiable reason. This occurred
more often for the ILDs, possibly due to calls being placed during loop retuning.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 13 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

Table 2. Summary of groundtruthed discrepancies.
(a) False Detection.
Discrepancy Classification
False Calls
Magnetometer “stuck on”
Magnetometer false call (adjacent
lane activity)
Magnetometer false call (unknown
reason)
ILD “stuck on”
ILD false call (adjacent lane
activity)
ILD false call (other reason)
Missed Calls
Motorcycle
Magnetometer
Other
missed call
Vehicle type
Motorcycle
Magnetometer
dropped call with
Other
redetection
Vehicle type
Motorcycle
Magnetometer
dropped call with
Other
no redetection
Vehicle type
Motorcycle
ILD missed call
Other
Vehicle type
Motorcycle
ILD dropped call
Other
with redetection
Vehicle type
Motorcycle
ILD dropped call
Other
with no redetection
Vehicle type

Northbound Left Turn
Southbound Left Turn
5/21 5/22 5/23 5/31 6/01 Total 5/24 5/25 5/26 5/27 5/28 Total
6

4

2

3

3

18

6

5

6

7

2

26

2

1

2

0

0

5

0

1

0

0

0

1

0

1

1

0

0

2

0

0

3

1

3

7

0

0

0

0

0

0

0

0

0

0

0

0

3

3

1

0

3

8

0

0

0

0

0

0

0

4

0

1

1

6

0

0

0

0

0

0

3

0

0

1

1

5

0

0

0

1

0

1

0

0

0

0

0

0

0

1

0

0

0

1

0

0

0

0

1

1

0

1

0

0

0

1

4

0

2

1

3

9

1

1

1

0

1

4

0

0

0

0

1

1

1

0

0

0

0

1

0

0

0

0

0

0

1

1

0

0

0

3

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

2

2

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

2

0

2

1

0

0

0

0

1

0

0

0

0

1

1

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

Missed and dropped calls are also tabulated for magnetometers and ILDs.


A missed call was defined as when a vehicle was not observed by the inactive detector.



A dropped call was defined as when a vehicle was detected upon its arrival, but the call
was dropped while the vehicle remained in the detection zone. If the vehicle was not
redetected (i.e., the detector stayed inactive) before it moved or another vehicle arrived at
the detection zone, the discrepancy was counted as a missed call.

These are further broken down by vehicle type, mostly because motorcycles were a common
source of missed and dropped calls, especially for the magnetometers. The magnetometers and

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 14 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

the ILDs missed or dropped a similar number of passenger cars. The low number of missed calls
suggested that there was no inherent tendency of the magnetometers to miss passenger cars with
the detector configuration shown in Figure 2b. For the ILDs, the dropped calls most likely
occurred due to detector card retuning (both misses took place within minutes of each other). We
speculate that some of the magnetometer misses might also have been due to short term detector
retuning, as some of those misses were nearly coincidental.
One possible explanation why the magnetometers missed more motorcycles than the ILDs is that
motorcycles tend to travel off-center in the lane and short vehicle length (in comparison to the
magnetometer spacing). In particular, it was observed that motorcyclists would often come to a
rest on the edge of an ILD, suggesting that there is an awareness that doing so would normally
actuate the phase (in the study period, the magnetometers were used for this purpose). Although
no bicycles were observed during any of the L1M0 type discrepancies, they would likely
encounter similar behavior.
Possibilities to improve detection of two wheel vehicles include positioning some
magnetometers slightly off-center in the lane, or including more than one toward the front of the
lane to achieve better longitudinal coverage. Another option might be to include pavement
marking indicating where vehicles should stop in order to actuate the signal. That solution may
be more effective with a laterally offset magnetometer rather than one positioned in the middle
of the grease stripe at the center of the lane. In locations where there is a wide paved shoulder
that vehicles may travel on, lateral coverage should be considered. At the test bed, although the
turn lanes were bounded by a grassy median, there were still a handful of vehicles that drove on
the shoulder and were consequently out of the range of the magnetometers.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 15 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

Activation and Termination Latency
As discussed in the previous section, distributions of magnetometer activation and termination
latency were computed. Figure 6 shows the histograms north and southbound left turns during
red/yellow and green phase with summary statistics. There are more activations during red and
more terminations in green because vehicles tend to arrive for left turn phases during red, and
depart on green.
In both activation and termination latencies, the distribution is slightly skewed to the right,
suggesting that the magnetometers tend to take more time than ILDs to decide whether to change
their reported state. The activation response of the ILDs and the magnetometers were nearly
synchronous, with 85th percentile magnetometer latencies being only a few tenths of a second.
Termination latencies tended to be somewhat longer.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 16 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

25%

20%
Relative Frequency

Rt,avg = 0.2
Rt,stdev = 0.3
Rt50% = 0.2
Rt85% = 0.3
Rt99% = 1.1
Rt100% = 2.1
n = 874

Ra,avg = 0.1
Ra,stdev = 0.2
Ra50% = 0.1
Ra85% = 0.2
Ra99% = 0.8
Ra100% = 2.1
n = 2658

15%

10%

5%

0%
-1

-0.5

0

0.5

1

1.5

2

-1

-0.5

Activation Latency (sec)

0

0.5

1

1.5

2

Termination Latency (sec)

NBLT during red and yellow intervals
25%

Ra,avg = 0.0
Ra,stdev = 0.1
Ra50% = 0.0
Ra85% = 0.1
Ra99% = 0.5
Ra100% = 1.2
n = 1080

Relative Frequency

20%

15%

10%

Rt,avg = 0.3
Rt,stdev = 0.3
Rt50% = 0.3
Rt85% = 0.5
Rt99% = 1.6
Rt100% = 2.8
n = 2890

5%

0%
-1

-0.5

0

0.5

1

1.5

2

-1

-0.5

Activation Latency (sec)

0

0.5

1

1.5

2

Termination Latency (sec)

NBLT during green intervals
25%

Ra,avg = 0.1
Ra,stdev = 0.2
Ra50% = 0.1
Ra85% = 0.2
Ra99% = 0.8
Ra100% = 2.6
n = 2664

Relative Frequency

20%

15%

10%

Rt,avg = 0.1
Rt,stdev = 0.2
Rt50% = 0.1
Rt85% = 0.3
Rt99% = 0.8
Rt100% = 2.7
n = 671

5%

0%
-1

-0.5

0

0.5

1

1.5

2

-1

-0.5

0

0.5

1

1.5

2

Termination Latency (sec)

Activation Latency (sec)

SBLT during red and yellow intervals
25%

Ra,avg = 0.1
Ra,stdev = 0.2
Ra50% = 0.0
Ra85% = 0.2
Ra99% = 0.6
Ra100% = 0.9
n = 589

Relative Frequency

20%

15%

10%

Rt,avg = 0.2
Rt,stdev = 0.3
Rt50% = 0.2
Rt85% = 0.4
Rt99% = 1.2
Rt100% = 2.7
n = 2652

5%

0%
-1

-0.5

0

0.5

1

1.5

2

-1

-0.5

0

0.5

1

1.5

2

Termination Latency (sec)

Activation Latency (sec)

SBLT during green intervals
Figure 6. Activation and termination latency by approach.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 17 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

Figure 7 offers another view of this information, the residual occupancy between the two
detector types:
Or  OM  O ILD ,

Equation 2

In this formula, Or, OM, and OILD respectively refer to residual, magnetometer, and ILD
occupancy time in seconds. The mean values of these distributions indicate that the
magnetometers report a slightly higher of occupancy than ILDs. This may be attributed to the
fact that the magnetometer termination latency is greater than the activation latency. For both
cases the variances are approximately equal to the mean indicating the distributions are normal.
We can then infer that there are no major causes of error in the calculations beyond typical slight
measurement stochasticity.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 18 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

25%

Mean = –0.09
St. Dev. = 0.37
Relative Frequency

20%

15%

10%

5%

0%
-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

1.5

2

Residual (sec)
(a) Northbound left turn (aggregated over 5 days).
25%

Mean = –0.08
St. Dev. = 0.37
Relative Frequency

20%

15%

10%

5%

0%
-2

-1.5

-1

-0.5

0

0.5

1

Residual (sec)
(b) Southbound left turn (aggregated over 5 days).

Figure 7. Occupancy residuals.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 19 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

Comparison of Magnetometer Performance to Agency Specification
Table 3 and Table 4 show a comparison of magnetometer performance statistics to the
specifications for standard performance shown in Table 1. Table 3 contains a summary of
activation and termination latency performance, while Table 4 provides a summary of missed
and false calls in terms of the worst case (24 hour and 1 hour periods) observed.
Table 3. Summary statistics for wireless magnetometer performance: Activation and termination latency
thresholds, aggregated over 120 hours.
Northbound Left Turn

Southbound Left Turn
Yellow and
Yellow and
Green
Green Interval Red
Red Intervals
Interval
Intervals

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 20 of 30

Pass

Threshold

Value

0.1  0.2 1.0  0.2 0.1 –
1.0  0.8 5.0  0.6 1.0 
1.0 – 2.6 5.0  0.9 1.0 
0.1 – 0.3 1.0  0.4 0.1 –
1.0 – 0.8 5.0  1.2 1.0 –
1.0 – 2.7 5.0  2.7 1.0 –

Pass

Threshold

0.1
0.5
1.2
0.5
1.6
2.8

Value








Pass

1.0
5.0
5.0
1.0
5.0
5.0

Threshold

Pass

0.2
0.8
2.1
0.2
1.1
2.1

Value

Threshold

Activation Response Time, Typical (Ra85%)
Activation Response Time, 99th percentile (Ra99%)
Activation Response Time, Maximum (Ra100%)
Termination Response Time, Typical (Rt85%)
Termination Response Time, 99th percentile (Rt99%)
Termination Response Time, Maximum (Rt100%)

Value

Test Parameter

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

Table 4. Summary statistics for wireless magnetometer performance: Worst-case observed numbers of
missed and false calls per detection zone.
Test Parameter
Number of Missed Calls, 24 hours (passenger cars)
Number of Missed Calls, 24 hours (motorcycles)
Number of Missed Calls, 24 hours (all vehicles)
Number of Missed Calls, 24 hours
Test Criteria
Pass
Number of Missed Calls, 1 hour (passenger cars)
Number of Missed Calls, 1 hour (motorcycles)
Number of Missed Calls, 1 hour (all vehicles)
Number of Missed Calls, 1 hour
Test Criteria
Pass
Number of False Calls, 24 hours
Number of False Calls, 24 hours
Test Criteria
Pass
Number of False Calls, 1 hour
Number of False Calls, 1 hour
Test Criteria
Pass
Magnetometers “stuck on,” 24 hours
1
No missed calls on 5/22/09 and 5/23/09.
2
No missed calls on 5/26/09 and 5/28/09.

Northbound Left Turn
Southbound Left Turn
Red/Yellow Green
Red/Yellow Green
Interval
Interval
Interval
Interval
0
0
2
0
3
0
1
0
3
0
2
0
0
10
0
10
(–)1

(–)2

0
0
2
0
1
0
1
0
1
0
2
0
0
10
0
10


(–)1
(–)2
8
9
20
20


3
2
20
20


6
7

The magnetometers satisfied () most of the specification requirements (or came very close to),
except for termination latency during green as reflected in Table 3. This is consistent with the
characteristics illustrated in Figure 6, where magnetometers had higher termination latencies than
activation latencies.
No missed calls were observed during the green intervals, indicating magnetometer performance
exceeded the specification requirements. The magnetometers passed () each requirement
except for the number of missed calls in a 24 hour period. The specification asks for zero missed
calls; however, as was noted in the discrepancy analysis (Three categories of false calls for each
detector technology were determined from video observation:


The active detector could become “stuck on,” holding the detect state for several seconds
after vehicles have left the detection zone. The operational impact was the same as that of

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 21 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

a false call. This was only observed for the magnetometers. This is discussed in more
detail later.


The active detector could present a false call due to adjacent lane activity, as confirmed
by lane incursions or the presence of wide vehicles. This occurred more often for the
ILDs than for the magnetometers. This is not unexpected because the ILDs are physically
more sensitive to lane incursions.



The active detector could present a false call without a verifiable reason. This occurred
more often for the ILDs, possibly due to calls being placed during loop retuning.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 22 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

Table 2), the magnetometers missed about the same number of passenger cars as the ILDs. The
(–) notation refers to the fact that there were several 24 hour periods in which no missed calls
took place. Consequently a (–) indication was used to signify that there were 24 hour periods
where the specification was met and 24 hour periods when the specification was not met.
As for false calls, the magnetometers exceeded the specification for both approaches. The
number of false calls does not include times when the magnetometers became “stuck on,” which
are shown in the last row of Table 4. This type of detection anomaly is not specifically addressed
in the specification. If included with false calls, the magnetometers would still exceed the
specified requirements.
In summary, the magnetometers passed most of the requirements of the detector specification.
The two requirements that were not achieved were 0.1 second termination latency in green and
zero missed calls.
Termination latency in green mainly affects the calculation of gaps to terminate phases. The
magnetometers provided 85th percentile green termination latencies of 0.4 seconds or less. The
effect of this latency would be to effectively add a few fractions of a second to gap times.

WIRELESS MAGNETOMETER DETECTION SYSTEM DESIGN
OBSERVATIONS AND RECOMMENDATIONS
During this evaluation, several lessons were learned with regard to magnetometer configuration
for stop bar detection. The existence of magnetometer “blind spots” can be problematic for
presence detection in areas where vehicles come to a stop. Figure 8a shows a plot of the change
in magnetic field density over time as a passenger car passes over a magnetometer. The initial
response is in the positive direction. However, as the vehicle passes by, there are commonly one
or more position where the frequency response crosses the zero line and becomes negative. If a
vehicle stops on or near a zero crossing, the vehicle is not detected. This phenomenon is not
observed in ILDs because their response parameter (relative change in inductance) can only have
positive values, as shown in Figure 8b. Magnetometer blind spots were found to lead to a large
number of missed calls when magnetometers were positioned with 15 ft spacing (Figure 2a), as
11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 23 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

determined by observation of many missed calls where vehicles were stopped at the same
approximate location. This led to the redesign of the detector configure shown in Figure 2b that
was used for conducting this evaluation.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 24 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

`
(a) Typical magnetometer response curve (16).

8

Loop Sensitivity L/L

7

6

5

4

3

2

1

0
0

500

1000

1500

2000

2500

3000

3500

time (ms)

(b) Typical ILD response curve (17).
Figure 8. Comparison of example typical detector response for a magnetometer and an ILD.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 25 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

The first phase of the testing performed in the spring of 2008 resulted in a substantial number of
dropped calls and detectors calls stuck on.


The stuck calls were attributed to the distance between the repeater and sensors slightly
exceeding the maximum design distance threshold of 150’. An additional RF Repeater
(Figure 1, RF Repeater III) was installed in October 2008 and corrected this problem.



The dropped calls were traced to the physics of the magnetometers where a vehicle
occasionally stops at a “zero crossing” similar to that shown in Figure 8a around 1600,
1700 and 1850 ms. With 15’ spacing between wireless magnetometers (Figure 2a), this
was a relatively common occurrence. Changing the spacing of the wireless
magnetometers from that shown in Figure 2a to the configuration shown in Figure 2b in
March 2009 eliminated the problem for passenger cars and greatly reduced it for
motorcycles.

Figure 9 shows typical ILD configurations for 3-detector (Figure 9a) and 4-detector (Figure 9b)
zones using 6 ft loops with spacing of 9 ft (typical INDOT detector configuration). Similar
detection zones can be designed with wireless magnetometers, as illustrated in Figure 9c. Four
magnetometers with a spacing of 8 ft are recommended for placement around the stop bar, with
additional detectors 12 ft apart to be placed further back in the lane to achieve the desired
detection zone length.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 26 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

3

2

6

9

1

6

9

6

36
(a) Three ILDs forming a 36 ft zone.

4

3

6

9

2

6

9

1

6

9

6

2

1

51
(b) Four ILDs forming a 51 ft zone.

6

5
12

4
12

3
8

8

8

36
48
(c) Recommended installation of wireless magnetometers for a 48 ft detection zone
Figure 9. Detection zone designs of various lengths.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 27 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

CONCLUSIONS
In summary, wireless magnetometer performance was evaluated in two left turn pockets by
comparing the outputs of co-located inductive loop detectors to identify discrepancies between
the two detection technologies. Discrepancies between the two technologies that exceeded three
seconds were visually groundtruthed to determine the cause. A comprehensive set of timestamped on-off detector data was collected over a 240 hour period. A summary of the missed and
false detections attributed to each detection technology is shown in Three categories of false calls
for each detector technology were determined from video observation:


The active detector could become “stuck on,” holding the detect state for several seconds
after vehicles have left the detection zone. The operational impact was the same as that of
a false call. This was only observed for the magnetometers. This is discussed in more
detail later.



The active detector could present a false call due to adjacent lane activity, as confirmed
by lane incursions or the presence of wide vehicles. This occurred more often for the
ILDs than for the magnetometers. This is not unexpected because the ILDs are physically
more sensitive to lane incursions.



The active detector could present a false call without a verifiable reason. This occurred
more often for the ILDs, possibly due to calls being placed during loop retuning.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 28 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

Table 2. Table 3 and Table 4 compare the observed results of the wireless magnetometers with
the target performance criteria defined in Table 1. Of particular note are the following:


Over the 10 days of testing, the loops and the wireless magnetometers had very similar
performance in terms of missed detections. Both sensors had complete 24 hour periods
with no missed calls. The loop detectors had 7 of 10 days with no dropped or missed
calls and the wireless magnetometers had 2 of 10 days with no dropped or missed
detections, but most of those were associated with intermittently dropping a call
associated with a motorcycle.



The wireless magnetometers had almost identical number of false calls (14 for loops and
15 for the wireless magnetometers). However, there were 44 events observed where the
wireless magnetometer “stuck on” until another vehicle arrived.

In regards to the activation/deactivation latency, Figure 6 shows the activation and deactivation
histograms during the green and red/yellow intervals for both the Northbound and Southbound
Left turn pocket. Histograms of nearly 6000 activation/deactivations demonstrated that there is a
time lag that is on average in the 0.1 to 0.3 second range. This latency is likely to a combination
of variation in detection zone location (see Figure 2b) and some minor communication latency.
However, the average occupancy residual shown in Figure 7 was observed to be less the 0.1
second for both turn lanes.
Although the wireless magnetometer installation shown in Figure 2b performed nearly
identically to loops, this particular spacing is an artifact of addressing performance problems
associated with 15’ sensor spacing encountered with the sensor spacing shown in Figure 2a.
Readers are reminded that these sensors have different footprints and responses that do not allow
one for one replacement of inductive loops with wireless magnetometers. The recommended
sensor spacing shown in Figure 9c represent engineering judgment regarding a more contractor
friendly sensor spacing that may potentially provide slightly better performance with
motorcycles at the stop bar.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 29 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

In conclusion, when the maximum sensor-repeater spacing is within manufacturer specifications,
and the sensor spacing shown in Figure 2b or Figure 9c is used, this detection technology
provides very similar performance to inductive loop detectors. However, agencies are cautioned
that they may need to implement pavement sensor tracking procedures to ensure that active
lithium thionyl chloride batteries are removed and disposed of in an environmentally friendly
manner prior to milling road surfaces or excavating pavements.
One direction for future research would be the development of design guidelines for adjusting
detection thresholds for magnetometers. Three-axis magnetometers have recently come to
market, which eliminates sensitivity of detector placement to positioning. In reality, only two
axes would be used (i.e., the sensitivity of one axis would be extremely low) due to the influence
of adjacent lane vehicles on one of them. A set of guidelines for establishing detector sensitivity
would be helpful in standardizing installations, as well as a procedures for recalibrating detection
zones over time.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 30 of 30

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

ACKNOWLEDGEMENTS
The wireless magnetometer used in this evaluation was produced by Sensys Networks, Inc.. Both
the preliminary design (Figure 2a) and the subsequent redesign (Figure 2b) of the detector
testbed were performed by Sensys. This study was supported by the Joint Transportation
Research Program administered by the Indiana Department of Transportation and Purdue
University. The contents of this paper reflect the views of the authors, who are responsible for
the facts and the accuracy of the data presented herein, and do not necessarily reflect the official
views or policies of the sponsors. These contents do not constitute a standard, specification, or
regulation.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 31 of 31

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

REFERENCES
1. Cheung, S.-Y., S. Coleri, B. Dundar, S. Ganesh, C.-W. Tan, and P. Varaiya. “Traffic Measurement and
Vehicle Classification With a Single Magnetic Sensor.” Transportation Research Record No. 1917, TRB,
National Research Council, Washington, DC, pp. 173-181, 2005.
2. Haoui, A., R. Kavaler, and P. Varaiya. “Wireless Magnetic Sensors for Traffic Surveillance.”
Transportation Research 16C, pp. 294-306, 2008.
3. Margulici, J.D., S. Yang, and C.-W. Tan. “Evaluation of Wireless Traffic Sensors by Sensys Networks,
Inc.” California Department of Transportation, 2006.
4. Cheung, S.-Y. and P. Varaiya. “Traffic Surveillance by Wireless Sensor Networks: Final Report.”
University of California PATH Research Report UCB-ITS-PRR-2007-4, 2007.
5. Abbas, M., and J. Bonneson. Video Detection for Intersection and Interchange Control. Publication
FHWA/TX-03/4285-1. Texas Transportation Institute, Texas Department of Transportation, FHWA,
2002.
6. MacCarley, C. A., and J. Palen. Evaluation of Video Traffic Sensors for Intersection Signal Actuation:
Methods and Metrics. Transportation Research Board, National Research Council, Washington, D.C.,
2003.
7. Rhodes, A., D. Bullock, J. Sturdevant, Z. Clark, and D. Candey, “Evaluation of Stop Bar Video
Detection Accuracy at Signalized Intersections," Transportation Research Record No. 1925, TRB,
National Research Council, Washington, DC, pp. 134-145, 2005.
8. Tian, Z., and M. Abbas. “Models for Quantitative Assessment of Video Detection System Impacts on
Signalized Intersection Operations.” Transportation Research Record No. 2035, TRB, National Research
Council, Washington, DC, pp. 50-58, 2007.
9. Rhodes, A., K. Jennings, and D. Bullock, “Consistencies of Video Detection Activation and Deactivation Times Between Day and Night Periods,” ASCE Journal of Transportation Engineering, Vol.
133, No. 9, pp. 505-512, September 2007.
10. Rhodes, A., E. Smaglik, D.M. Bullock and J. Sturdevant, “Operational Performance Comparison of
Video Detection Systems,” Proceedings of the 2007 ITE International Annual Meeting and Exhibit,
August 5-8, 2007.
11. Middleton, D., R. Longmire, D.M. Bullock, and J.R. Sturdevant. “A Proposed Concept for Specifying
Vehicle Detection Performance,” forthcoming in Transportation Research Record, Paper No. 09-3013,
2009.
12. Medina, J.C., M.V. Chitturi, and R.F. Benekohal. “Illumination and Wind Effects on Video Detection
Performance at Signalized Intersections.” Transportation Research Board 87th Annual Meeting DVD,
Paper No. 08-2866, TRB, National Research Council, Washington, DC, 2008.
13. Medina, J.C., M.V. Chitturi, and R.F. Benekohal. “Impact of Snow on Video Detection Systems at
Signalized Intersections.” Transportation Research E-Circular No. E-C126, TRB, National Research
Council, Washington, DC, pp. 469-479, 2008.
14. Medina, J.C., R.F. Benekohal, and M.V. Chitturi. “Changes in Video Detection Performance at
Signalized Intersections Over Different Illumination Conditions.” Transportation Research Board 88th
Annual Meeting DVD, Paper No. 09-3595, TRB, National Research Council, Washington, DC, 2009.
15. Indiana Department of Transportation,” Procedure for Evaluating Vehicle Detection Performance,”
ITM No. 934-08P, January 2008 (web accessible at
http://www.in.gov/indot/div/M&T/itm/pubs/934_testing.pdf).
11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 32 of 32

1:37:40 PM
Paper revised from original submittal.

Day, Premachandra, Brennan, Sturdevant, and Bullock

16. Ernst, J.M., J.V. Krogmeier, A. Ault, and D.M. Bullock,” Recommended Tolerances for
Magnetometer Orientation and Field Calibration Procedure,” Annual Meeting DVD, Paper No. 09-0202,
TRB, National Research Council, Washington, DC, 2009.
17. Day, C.M., T.M. Brennan, M.L. Harding, H. Premachandra, A. Jacobs, D.M. Bullock, J.V.
Krogmeier, and J.R. Sturdevant. “Three Dimensional Mapping of Inductive Loop Detector Sensitivity
Using Field Measurement,” Annual Meeting DVD, Paper No. 09-0018, TRB, National Research Council,
Washington, DC, 2009.

11/13/2009
TRB 2010 Annual Meeting CD-ROM

Page 33 of 33

1:37:40 PM
Paper revised from original submittal.


10-0007.pdf - page 1/33
 
10-0007.pdf - page 2/33
10-0007.pdf - page 3/33
10-0007.pdf - page 4/33
10-0007.pdf - page 5/33
10-0007.pdf - page 6/33
 




Télécharger le fichier (PDF)

10-0007.pdf (PDF, 443 Ko)

Télécharger
Formats alternatifs: ZIP



Documents similaires


10 0007
10 0008
bed bug detecting canines 1
fichier pdf sans nom
esophageal ph monitoring and impedance 9 1
physreva 84 020303

Sur le même sujet..