ITS - Intelligent Transportation Systems Report ITS Home Page

4. Using Toll Tag Readers for Traffic Monitoring

One of the innovative parts of the iFlorida design was the extensive use of toll tag readers to collect vehicle probe data for estimating travel times and identifying incidents. (The system also used a limited number of license plate readers. The focus in this report will be on the toll tag reader network, though many of the observations would apply equally well to a network of license plate readers.) This section of the report describes the iFlorida toll tag travel time system, the challenges faced by FDOT in deploying, operating, and maintaining this system, and the success of the system in producing travel times. A particular focus is given to the characteristics of the arterials on which the travel time system was deployed and the impact of those characteristics on the effectiveness of the travel time system.

4.1. Using Toll Tag Reads to Estimate Travel Times

Toll tag reads were used to estimate travel times for a segment of road by matching tag numbers read at the starting point of the segment of road with tag numbers read at the ending point and estimating the travel time as the difference between the times when two matching tag reads were made. In this sense, a travel time determined by matching toll tag readings was a direct measure of the travel time taken by a vehicle that actually traveled the segment in question. If the system identified every vehicle entering and exiting a segment and if each vehicle drove without diversion from the start to the end of the segment, then this approach would completely characterize the travels times experienced by vehicles on the segment.

In practice, the performance of the system will be much less for many reasons - not all vehicles will be equipped with toll tags, not all toll tags will be read, etc. The remainder of this section describes the process for computing travel times from toll tag reads and identifies different types of failures that might occur during this process. Later sections of the document give examples of these failures that occurred during the iFlorida deployment and describes the effectiveness of the iFlorida toll tag travel time system. Figure 26 depicts the process for estimating travel times from toll tag readings, calling out the factors that can reduce the effectiveness of the toll tag travel time estimation process.

Image is a flow chart depicting the process for estimating travel times from toll tag readings and calls out the factors that can reduce the effectiveness of the toll tag travel time estimation process.
Figure 26. The Process for Estimating Travel Times from Toll Tag Readings

Each of these factors is described in the list below.

The effectiveness of a toll tag travel time system may be limited by a number of factors, including the fraction of vehicles equipped with toll tags, the effectiveness of the toll tag readers, duplicate reads, vehicle diversions and stops, and mis-matched tag reads.

Other potential failure modes were associated with the specific process used in the iFlorida deployment to estimate and make use of travel times from toll tag readings (see Figure 27).

Flow diagram summarizes the key steps in the travel time process and what activities take place at each step of information development.
Figure 27. The iFlorida Toll Tag Travel Time Process

A toll tag travel time system should accommodate many different types of failures that might occur, including reader failures, clock mis-synchronization, incorrect reader location assignments, and network failures.

This process worked as follows. A toll tag was read from a vehicle passing one of the iFlorida toll tag readers and the data from that read was transmitted to a roadside central processing unit (CPU). This CPU accumulated readings from all readers deployed at a single site, archived those reads onto a local hard disk, and transmited each read to a central server. The roadside CPU also tracked transmissions that failed to reach the central server so that it ccould retransmit them at a future time. The central server batched all of the reads received during each one-minute interval, archived these one-minute batches, and transmited them to the travel time server. The travel time server combined each batch of data with other data previously received, matched toll tag reads across links to produce travel time measurements, excluded outliers from these measurements, and calculated the average travel time from the remaining measurements. The resulting travel time measurements were transmitted to the RTMC software, which used the travel times to support traffic management activities (e.g., produce automated 511 travel time messages) and archived the travel times in a data warehouse. The following list summarizes some failures that occurred as part of this specific toll tag travel time process.

4.2. The iFlorida Toll Tag Travel Time Network

The development of the iFlorida toll tag travel time network began with studies to test the feasibility of using toll tag reads to measure travel times on Florida roads. One study1 reported on field tests conducted on Florida Interstate highways. Field tests were conducted on urban portions of I-4 and I-95 and on rural portions of I-10, I-75, and I-95. These tests indicated that toll tag penetration was high in urban areas with toll roads, but was much lower in other areas:

Field studies can help determine if toll tag market penetration is sufficient to support the use of a toll tag travel time system.

Additional tests were conducted on Orlando arterials.2 These tests showed a similar level of toll tag penetration in vehicles on Orlando arterials as observed on I-4 - toll tag reads were made for 15 to 20 percent of vehicles at the eight observation sites considered in the study. This report also identified the following information about the percent of reads at a station for which a matching read was identified at a second station, so that a travel time estimate could be made.3

For the segments considered in this study, the toll tag readers generated between 10 and 20 travel time estimates per hour. Based on this information about the feasibility of using toll tag readers to estimate travel times on Orlando arterials and limited access roads, FDOT developed a plan for doing so. The locations of the toll tag readers is depicted in Figure 3 on page 5, with the resulting travel time segments listed in Table 2.

Table 2. iFlorida Arterial Travel Time Segments
Road Segment Length (miles) Lanes Major Intersections Volume
SR 50 (Colonial Dr) SR 91 to SR 429 4.9 2 7 28,500 to 50,000
SR 50 (Colonial Dr) SR 429 to SR 408 2.4 2 4 56,000 to 44,500
SR 50 (Colonial Dr) SR 408 to SR 423 6.5 3 17 33,500 to 47,500
SR 50 (Colonial Dr) SR 423 to SR 441 1.1 3 3 47,500 to 36,500
SR 50 (Colonial Dr) SR 441 to I-4 0.8 2 3 41,500 to 44,500
SR 50 (Colonial Dr) I-4 to US 17/92 1 2 4 44,500 to 43,000
SR 50 (Colonial Dr) US 17/92 to SR 436 3.4 3 12 48,000 to 65,000
SR 50 (Colonial Dr) SR 436 to SR 417 3 3 4 45,500 to 44,500
SR 50 (Colonial Dr) SR 417 to SR 408 4.5 3 8 53,500 to 45,500
SR 414 (Maitland Blvd) SR 441 to I-4 4.3 3 2 31,000 to 66,500
SR 414 (Maitland Blvd) I-4 to US 17/92 2 2 7 56,000 to 31,500
SR 423 (John Young Parkway) (Lee Rd) SR 417 to SR 528 2.8 3 3 (no data)
SR 423 (John Young Parkway) (Lee Rd) SR 528 to I-4 6.1 3 3 (no data)
SR 423 (John Young Parkway) (Lee Rd) I-4 to SR 408 3 3 3 (no data)
SR 423 (John Young Parkway) (Lee Rd) SR 408 to SR 50 0.7 3 3 48,000 to 49,000
SR 423 (John Young Parkway) (Lee Rd) SR 50 to SR 441 3.2 2 2 46,500 to 40,500
SR 423 (John Young Parkway) (Lee Rd) SR 441 to I-4 2.1 3 3 38,500 to 44,500
SR 423 (John Young Parkway) (Lee Rd) I-4 to US 17/92 1.3 2 2 40,500
SR 436 (Semoran Blvd) SR 441 to SR 434 5.4 4 12 23,000 to 48,500
SR 436 (Semoran Blvd) SR 434 to I-4 1.8 4 7 48,500
SR 436 (Semoran Blvd) I-4 to US 17/92 3.1 4 12 63,000 to 62,000
SR 436 (Semoran Blvd) US 17/92 to SR 50 7.9 3 19 70,500 to 52,000
SR 436 (Semoran Blvd) SR 50 to SR 408 1.2 3 3 55,000 to 52,500
SR 436 (Semoran Blvd) SR 408 to SR 528 5.4 3 13 57,000 to 42,000
SR 441 (Orange Blossom Trail) US 192 to SR 417 4.7 3 7 30,500 to 34,500
SR 441 (Orange Blossom Trail) SR 417 to SR 528 4.4 3 8 50,500 to 42,000
SR 441 (Orange Blossom Trail) SR 528 to I-4 4.7 3 13 61,500 to 66,500
SR 441 (Orange Blossom Trail) I-4 to SR 408 2.2 2 5 36,500
SR 441 (Orange Blossom Trail) SR 408 to SR 50 1.2 2 7 36,500 to 28,500
SR 441 (Orange Blossom Trail) SR 50 to SR 423 3.4 2 4 32,500 to 31,000
SR 441 (Orange Blossom Trail) SR 423 to SR 414 3.4 2 4 35,000 to 33,000
SR 441 (Orange Blossom Trail) SR 414 to SR 436 3.5 2 5 33,000 to 27,000
SR 441 (Orange Blossom Trail) SR 436 to SR 429 2.2 2 4 41,500 to 38,000
US 192 (West Coast Parkway) I-4 to SR 441 9.8 3 18 54,000 to 37,000
US 192 (West Coast Parkway) SR 441 to SR 91 3.6 3 9 35,500 to 36,000
US 17/92 (Orlando Ave) SR 50 to SR 423 2.7 2 9 28,000 to 37,500
US 17/92 (Orlando Ave) SR 423 to SR 414 2.3 3 4 45,500 to 33,500
US 17/92 (Orlando Ave) SR 414 to SR 436 1.9 3 2 61,000 to 55,500
US 17/92 (Orlando Ave) SR 436 to SR 417 8.8 3 15 60,500 to 43,500

In this table, the Road column identifies the road on which each travel time segment lies with the segments, running either from East to West or from South to North, identified by the cross-streets bounding each segment. The length of the segment and the number of lanes was provided by FDOT. The number of intersections column lists the number of major intersections within each segment (i.e., excluding the intersections at the start and end of each segment), where each interchange and each intersection with a stop bar was identified as a major intersection. The volume information was obtained from FDOT statewide traffic count.4

Table 3 lists similar information for the FDOT toll tag travel time segments on limited access highways and on evacuation routes where travel time were monitored by either toll tag or license plate readers.

Table 3. iFlorida Limited Access and Evacuation Travel Time Segments
Road Segment Length (miles) Lanes Major Intersections Volume
SR 91 (The Florida Turnpike) US 192 to SR 522 4.2 2 0 32,800
SR 91 (The Florida Turnpike) SR 522 to SR 528 6.9 2 0 32,800 to 41,700
SR 91 (The Florida Turnpike) SR 528 to I-4 4.1 2 0 41,700 to 44,700
SR 91 (The Florida Turnpike) I-4 to SR 408 6.1 2 0 44,700
SR 91 (The Florida Turnpike) SR 408 to SR 50 7 2 1 44,700 to 48,100
SR 417 I-4 to Boundary 5.1 2 2 10,200 to 16,300
SR 417 Boundary to SR 434 6.1 2 2 45,000 to 28,800
SR 417 SR 434 to US 17/92 6.6 2 1 24,100 to 14,500
SR 417 US 17/92 to I-4 4.8 2 2 (no data)
SR 520 SR 528 to I-95 13.9 1 2 (no data)
SR 528 I-4 to SR 91 4.3 2 3 60,600 to 54,400
SR 528 SR 91 to Boundary 4.6 2 0 54,400 to 49,800



4.3. The iFlorida Toll Tag Travel Time Estimation Process

As each toll tag reader observed toll tag IDs on passing vehicles, it transmited an encrypted hash key representing that ID along with the time it was observed to a Travel Time Server. This server matched toll tag observations made at readers at the start and end of each travel time segment to generate travel time measurements. Anomalous measurements were discarded, and the remaining measurements were used to estimate the average travel time, using one of two possible approaches. In the first approach, a rolling average of recent observations over a fixed period of time was used. This approach was used when travel time observations were plentiful. In the second approach, a rolling average of the most recent fixed number of observations was used. This approach was used when fewer observations were available. Within this general framework, a number of configurable parameters were used to control the specific operations of the process.

This travel time process was originally designed for computing travel times for OOCEA on limited access toll highways, and the configuration parameters were set to the default values used on those highways. A number of adjustments were made to try to optimize the parameters for use on Orlando arterials.

4.4. An Example – Toll Tag Travel Time Estimates on Colonial Drive

This section exemplifies the performance of the toll tag travel time system by examining its performance on the eastbound section of Colonial Drive between US 17/92 (Mills Avenue) and SR 436 (Semoran Boulevard). This helps lay the groundwork for a more general analysis of the entire network by identifying performance characteristics that should be considered in the more general analysis.

Table 2 indicates that this section of road is 3.4 miles long, has 3 lanes in each direction of travel, includes 12 major intersections, and has a traffic volume ranging from 48,000 vehicles per day at the western portion of the section to 65,000 vehicles per day at the eastern portion of the section. The configuration of the intersection at the start of this travel time link is depicted in Figure 28.

Diagram of a four-way intersection with icons indicating the locations of toll tag readers at each of the four stop approaches. Numbers at each directional indicate traffic counts.
Figure 28. The Intersection of SR 50 and US 17/92

In this figure, the numbers are intersection traffic counts from City of Orlando traffic count program and the icons beside the roads represent the approximate location of the toll tag readers. Figure 29 depicts the configuration of the intersection at the end of this travel time link.

Diagram of a four-way intersection that is not at right angles, but rather in the shape of an X that contains icons indicating the locations of toll tag readers at each of the four stop approaches.
Figure 29. The Intersection of SR 50 and SR 436

Not shown in this figure is a road leading from SR 50 west of SR 436 to SR 436 southbound, which serves as a de facto turn lane for eastbound traffic on SR 50 wishing to turn right onto SR 436.

The iFlorida travel time process for this travel time segment matched reads from the toll tag reader monitoring eastbound traffic on SR 50 east of the US 17/92 intersection with those from the reader monitoring eastbound traffic on SR 50 east of SR 436. (An alternate approach would be to include reads from the three readers at the SR 50 / SR 436 intersection monitoring eastbound traffic on SR 50 and north- and south-bound traffic on SR 436, though this approach would mix travel times that did and did not include the additional delay associated with making a left turn from SR 50 onto SR 436.) The number of tag reads at the SR 50 Eastbound, East of US 17/92 reader are depicted in Figure 30 (for weekdays) and Figure 31 (for weekends).

Line graph showing the number of tag reads at the SR 50 Eastbound on weekdays. The number of reads per day remained between 4,000 and 6,300 on most weekdays.
Figure 30. Weekday Tag Reads at the SR 50 Eastbound, East of US 17/92 Reader

Line graph showing the number of tag reads at the SR 50 Eastbound on weekends. The number of reads per day were between 2,500 and 5,000 on most weekend days.
Figure 31. Weekend Tag Reads at the SR 50 Eastbound, East of US 17/92 Reader

Note that this reader first began operating stably in December 2006. Prior to that time, the reader often had no tag reads or fewer reads than should have been expected. Once it began operating stably, the number of reads per day remained between 4,000 and 6,300 on most weekdays and between 2,500 and 5,000 on most weekend days. The period in early August 2006 when more reads than usual were observed occurred when the reader re-transmitted the toll tag reads that occurred on August 1 and 2, 2006. After December 2006, the reader operated with only four days of very few reads and nine days with fewer reads than expected, but still sufficiently many to support travel time calculations.

Figure 32 is a frequency chart listing the number of days the reader at SR 50 Eastbound East of US 17/92 had a number of reads within a series of ranges for the period from 1/1/2007 through 8/26/2007. The lowest range category was 0 to 2,000 reads and the highest range category was more than 6,500 reads, with the size of each category in between being 250 reads.

Frequency chart listing the number of days the reader at SR 50 Eastbound had a number of reads within a series of ranges for a given period. The lowest range category was 0 to 2,000 reads and the highest range category was more than 6,500 reads, with the size of each category in between being 250 reads. Graph indicates that the eastbound volume at the reader location was about 21,000 vehicles per day.
Figure 32. Toll Tag Reads at SR 50 Eastbound, East of US 17/92, by Day of Week

This indicates that the range for the number of reads per day was similar across weekdays, but different for Saturday and Sunday. Note that the average number of reads per day was about 4,500. Of these, about 500 reads per day were duplicate reads - multiple reads of the same tag at the same location within a five minute period - leaving about 4,000 vehicles read per day. Figure 28 indicates that the eastbound volume at the reader location was about 21,000 vehicles per day, indicating that the reader was identifying about 19 percent of the passing vehicles. This was consistent with the observations made by FDOT prior to the iFlorida deployment.

Another measure of the operational status of a reader for supporting travel time calculations was the lag time between when a toll tag read was made and when it reached the toll tag server. Figure 33 depicts the measured lag time for the SR 50 Eastbound, East of US 17/92 reader.

Chart shows the measured lag time between when toll tag read was made and when it reached the toll tag server for the SR 50 Eastbound East of US 17/92 reader. Lag is in excess of 30 days at several points on the chart.
Figure 33. Lag Time for the SR 50 Eastbound, East of US 17/92 Reader

Because the lag time was measured as the difference between when a reading was timestamped by a roadside CPU and when it was logged at the AVI server, there were several possible causes for a lag:

Based on the number of reads and the lag time, a value representing the operational status of the reader was assigned to the reader for each day. The following codes were used:

The following four figures depict the operational status of each toll tag reader associated with the travel time segment on SR 50 from US 17/92 to SR 436. Note that a maximum value of 0 or 1 indicates that the reader was 100 percent operational from the standpoint of computing travel times, a value of 2 indicates that the reader was operating at lower effectiveness, and that higher values indicate that the reader could not be used to provide real-time travel times.

Graph displays operational status of toll tag reader at SR 50 from US 17/92 to SR 436. From February 2005 through January 2007, readers were rated at 3 or above for a significant period of time.
Figure 34. Errors at the SR 50 Eastbound, East of US 17/92 AVI Reader

Graph displays operational status of toll tag reader at SR 50 Eastbound, East of SR 436 AVI. The period from February 2005 through August 2005 showed that the reader was largely not operational, although periods of inoperability continued sporadically through July of 2007.
Figure 35. Errors at the SR 50 Eastbound, East of SR 436 AVI Reader

Graph displays operational status of toll tag reader at the SR 436 Northbound, North of SR 50 AVI. The period from February 2005 through August 2005 showed that the reader was largely not operational, although periods of inoperability continued sporadically through July of 2007.
Figure 36. Errors at the SR 436 Northbound, North of SR 50 AVI Reader

Graph displays operational status of toll tag reader at SR 436 Southbound, South of SR 50 AVI. The period from February 2005 through August 2005 showed that the reader was largely not operational, although periods of inoperability continued sporadically through July of 2007.
Figure 37. Errors at the SR 436 Southbound, South of SR 50 AVI Reader

Figure 38 combines the error information for the above four readers to identify the operational status of the travel time system for the travel time segment, with a value of 100 percent being used if the three readers used for the travel time calculation were all fully operational and lower percentages being used to represent degraded performance if one or more of the readers was exhibiting one or more deficiencies.

Graph displays combined operational status of all toll tag readers for SR 50 from US 17/92 to SR 436. The period from February 2005 through January 2007 shows frequent periods of zero operability, with sporadic periods where more than 50 percent of readers were operable. From January 2007 through July 2007, however, the readers were largely in the 100 percent operability range, with only sporadic periods where fewer than 50 percent were operable.
Figure 38. Status of the Travel Time System for SR 50 from US 17/92 to SR 436

Figure 39 summarizes this data by computing the average percent operational for each month.

Graph displays a summary of the travel time system for SR 50 from US 17/92 to SR 436 by month from March 2005 through July 2007. Data indicate that, prior to 2007, the network was unreliable. Over the first eight months of 2007, the system was above 90 percent operational in four months, with lower levels of reliability in the other months.
Figure 39. Monthly Status of the Travel Time System for SR 50 from US 17/92 to SR 436

Together, the above charts and figures provide measures for assessing the operational status of the portion of the toll tag reader network needed to support travel time estimates for the travel time segment on SR 50 from US 17/92 to SR 436. They indicate that, prior to 2007, the network was unreliable. Over the first eight months of 2007, the system was above 90 percent operational in four months, with lower levels of reliability in the other months.

But, the operational status of the toll tag reader only tells part of the story of a toll tag travel time network. Because the iFlorida arterial travel time segments were long with many opportunities for travelers to divert, it was possible that many travelers entering the segment did not exit it at its endpoint. The percentage of (non-duplicated) toll tag reads for vehicles entering the travel time segment during the period from August 13, 2007 to August 19, 2007 are listed in Table 4. (That time period was chosen because the readers for this travel time segment did not exhibit any deficiencies on those days.)

Table 4. Matching Effectiveness for the SR 50 from US 17/92 to SR 436 Segment
Date Duplicate Reads Non-Matching Reads Matching Reads Matching Percent
8/13/2007 345 3,165 612 16.20%
8/14/2007 319 3,262 610 15.80%
8/15/2007 421 3,265 692 17.50%
8/16/2007 373 3,350 690 17.10%
8/17/2007 423 3,350 598 15.10%
8/18/2007 286 2,787 496 15.10%
8/19/2007 149 2,321 440 15.90%

This table indicates that, for this travel time segment, about 16 percent of the toll tag reads made at the start of the segment would result in travel time measurements. However, some of these measurements may be invalid. For example, if a traveler makes a stop mid-segment, then later completes the segment, the result would be a travel time measurement that included the time the traveler spent stopped. This was more problematic for arterials than limited access highways because stops were more likely to occur and, if a stop occurred, it was more difficult to differentiate between a stop-induced delay and a delay induced by traffic signals encountered while traveling the segment. Figure 40 depicts the distribution for the travel times observed during the period from Monday August 13, 2007 through Friday August 17, 2007.

Graph displays combined travel times from 0 to 60 minutes for a 24 hour period in multi-hour time segments. While the fraction of measurement for 0 to 5 minutes and 15 to 60 minutes is nearly zero, travel times of 5 minutes to 15 minutes show fractions of measurement ranging from 0.025 up to more than 0.350.
Figure 40. Travel Time Distribution for the SR 50 from US 17/92 to SR 436 Segment

The important thing to note about these distributions is the long tail of high travel times, indicative of travel times that include time spent stopped or diverting. The effect of these outliers on the mean travel time is shown in Table 5.

The effect of diverting traffic on arterials can introduce a high bias into the observed average travel time.

Table 5. Travel Time Statistics for the SR 50 from US 17/92 to SR 436 Segment
Time Period Count Median Mean Percent Outlier
Midnight to 6:00 am 257 6.1 7.4 5%
6:00 am to 9:00 am 263 8.8 10.3 5%
9:00 am to 11:00 am 319 9 14 18%
11:00 am to 1:00 pm 375 12.2 22.5 25%
1:00 pm to 4:00 pm 665 11.5 20.9 23%
4:00 pm to 7:00 pm 700 11.8 19.4 19%
7:00 pm to 11:00 pm 533 10.4 17.6 17%

In this table, the Percent Outlier column indicates the percentage of high travel time measurements that should be discarded as outliers in order for the corrected mean (i.e., mean for measurements after outliers are removed) to equal the median. Note that, without correcting for the outliers, the mean is a poor estimator of the travel time experienced by most travelers. Based on this example, it appears that the median may be a better choice for estimating the segment travel time based on arterial probe measurements than the mean. If the mean is used, a method for removing high travel time outliers resulting from intermediate stops and diversions must be used.

For limited access highways, diverting traffic is less likely to introduce a high bias into the observed average travel time.

For comparison, Figure 41 and Table 6 show corresponding travel times for a limited access section of highway, SR 417 from I-4 north of Orlando to US 17/92.

Graph depicts travel times from 0 to 60 minutes for a 24 hour period in multi-hour time segments. Outlier points for all time segments occur at approximately the 2 to 5 minute travel time poiint, with fractions of measurement for each segment ranging from 0.025 up to more than 0.700.
Figure 41. Travel Time Distribution for the SR 417 from I-4 to US 17/92 Segment

Table 6. Travel Time Statistics for the SR 417 from I-4 to US 17/92 Segment
Time Period Count Median Mean
Midnight to 6:00 am 153 4.1 4.7
6:00 am to 9:00 am 869 4 4.3
9:00 am to 11:00 am 307 4.1 4.7
11:00 am to 1:00 pm 218 4.1 5.2
1:00 pm to 4:00 pm 441 4.1 5.3
4:00 pm to 7:00 pm 751 4.2 4.8
7:00 pm to 11:00 pm 298 4.1 4.7

Note that, for the limited access highway, outliers with travel time significantly above the median value were much less common and the mean and median were relatively close to each other in value.

Another question related to the effectiveness of a toll tag travel time system was whether it will generate enough toll tag matches to produce reliable travel times. Figure 42 depicts the average number of travel time reads per hour for the SR 417 from I-4 to US 17/92 segment during the period from August 13, 2007 to August 17, 2007.

Graph depicts the average number of travel time reads per hour for the SR 417 from I-4 to US 17/92 segment during the period from August 13, 2007 to August 17, 2007, with the total number of reads broken out by time of day. Reads typically increase from about 15 reads at 6 a.m. on, with the peak occurring at just over 50 reads per hour at about 4 p.m., then declining steadily throughout the remainder of the day, ending at about 19 reads just before midnight.
Figure 42. Travel Times Per Hour for the for the SR 417 from I-4 to US 17/92 Segment

The following approach can be used to generate a rough estimate of the number of travel time measurements needed to generate an accurate estimate for the average travel time. The curves in Figure 40 indicate that the width of the travel time distributions is roughly equal to the median, which is roughly equal to the mean after the top 25 percent of measurements are removed as outliers. Under these circumstances, a sample size of 25 (after removing outliers) would result in an estimate of the mean travel time that should be within 15 percent of the actual value about 90 percent of the time. A sample size of about 32 would provide enough measurements so that, after removing outliers, a sample size of 25 would remain. During peak hours, it required more than 30 minutes to accumulate 32 travel time measurements. During night time hours, it required about 3 hours to accumulate 32 travel time measurements. On weekends, when traffic volumes are lower, longer accumulation times would be needed.

These observations pointed to a tradeoff that must be considered when designing an arterial toll tag travel time system. If one assumes that the travel times do not change significantly over time, then averaging over a long time period will increase the number of observations over which the average is taken, which increases the statistical accuracy of the estimate for the mean travel time. If travel times do change over time, averaging over a long period of time would include in the average travel times that no longer represent current travel time conditions. This is exemplified in Figure 43, which depicts travel times on the SR 50 from US 17/92 to SR 436 travel time segment during the period from 6:00 pm to 7:00 pm on August 13, 2007, a time during which travel times were decreasing as rush hour ended. (Travel times greater than 20 minutes have been excluded from this chart.)

Scatter graph depicts travel times on the SR 50 from US 17/92 to SR 436. During the period from 6:00 pm to 7:00 pm on August 13, 2007, travel times were decreasing as rush hour ended.
Figure 43. Travel Times for SR 50 from US 17/92 to SR 436 on 8/13/2007

Figure 44 shows the average travel times computed from the observed travel times on this day, using different values for the length of the interval (i.e., the lag time) over which the average was computed.

Graph shows the average travel times computed from the observed travel times on August 13, 2007, using different values for the length of the interval (i.e., the lag time) over which the average was computed. As the lag time increased, the travel time curve was smoother, but the average travel time was larger than the recently observed travel times.
Figure 44. Average Travel Times for SR 50 from US 17/92 to SR 436 on 8/13/2007

Note that, as the lag time increased, the travel time curve was smoother, but the average travel time was larger than the recently observed travel times because the average included observed travel times from further in the past when congestion levels were higher. This is generally true. Averaging over short periods of time will result in highly variable estimates for the average travel time because the averages include few observations. The situation would be much worse if traffic conditions changed more abruptly, as might occur if a crash occurred on the travel time segment.

The main problem was that there are relatively few observations - averaging about one observation every two minutes for the period in question. Averaging over a short period of time, so that the estimated travel times responded quickly to changes in traffic conditions, meant that the estimated travel times were based on very few measurements and the variability in the estimated travel times was high. Averaging over a longer period of time to reduce this variability dampened the response to changes in traffic conditions.

One way to improve the performance of the travel time system would be to increase the number of observations by allowing multiple toll tag readers to supply measurements at the end point of a travel time segment. In the example at hand, all vehicles entering the segment pass by the toll tag reader on SR 50 eastbound, east of US 17/92. Vehicles that turn onto SR 436 at the end of the segment do not pass by the reader on SR 50 eastbound, east of SR 436 - left-turning vehicle pass by the reader on SR 436 northbound, north of SR 50 and most right-turning vehicles do not pass by a reader because the reader location is north of the entry point taken by most right-turning vehicles. The Travel Time Server used for iFlorida only allows use of a single reader for the endpoint of a travel time segment. A tool developed by the Evaluation Team allowed multiple readers to be used at both the start and end point of a travel time segment. Table 7 summarizes results travel time results for the section of SR 50 from US 17/92 to SR 436 using different numbers of readers for the end node.

Table 7. Travel Time Statistics for the SR 50 from US 17/92 to SR 436 with Different End Node Configurations
End Node Time Period Raw Data Corrected Data
Match Count Match Percent Mean TT (min) Match Count Match Percent Mean
SR 50 Eastbound and SR 436 Northbound Midnight to 6:00 am 385 34% 8.3 273 24% 6
SR 50 Eastbound and SR 436 Northbound 6:00 am to 9:00 am 394 19% 11.1 314 15% 8.4
SR 50 Eastbound and SR 436 Northbound 9:00 am to 11:00 am 394 17% 12.8 329 14% 9.4
SR 50 Eastbound and SR 436 Northbound 11:00 am to 1:00 pm 449 13% 16.9 345 10% 11.6
SR 50 Eastbound and SR 436 Northbound 1:00 pm to 4:00 pm 831 16% 19 659 13% 11.6
SR 50 Eastbound and SR 436 Northbound 4:00 pm to 7:00 pm 925 19% 17.1 763 16% 11.5
SR 50 Eastbound and SR 436 Northbound 7:00 pm to 11:00 pm 710 21% 16 574 17% 10.3
SR 50 Eastbound Only Midnight to 6:00 am 265 24% 8.2 168 15% 5.8
SR 50 Eastbound Only 6:00 am to 9:00 am 267 13% 10.8 209 10% 8.5
SR 50 Eastbound Only 9:00 am to 11:00 am 284 12% 12.5 231 10% 9.2
SR 50 Eastbound Only 11:00 am to 1:00 pm 326 9% 17.1 249 7% 11.5
SR 50 Eastbound Only 1:00 pm to 4:00 pm 609 12% 18.2 481 10% 11.3
SR 50 Eastbound Only 4:00 pm to 7:00 pm 646 13% 16.8 522 11% 11.5
SR 50 Eastbound Only 7:00 pm to 11:00 pm 506 15% 16.1 398 12% 10.4
SR 436 Northbound Only Midnight to 6:00 am 151 13% 9 80 7% 6.2
SR 436 Northbound Only 6:00 am to 9:00 am 144 7% 12.3 101 5% 8.5
SR 436 Northbound Only 9:00 am to 11:00 am 122 5% 14.4 90 4% 10
SR 436 Northbound Only 11:00 am to 1:00 pm 149 4% 17.9 106 3% 12.8
SR 436 Northbound Only 1:00 pm to 4:00 pm 268 5% 21.8 199 4% 12.6
SR 436 Northbound Only 4:00 pm to 7:00 pm 333 7% 18.5 260 5% 11.5
SR 436 Northbound Only 7:00 pm to 11:00 pm 251 7% 16.4 179 5% 10.7

In this table, the corrected data was generated by excluding from the mean those travel time values that were more than 0.5 standard deviations above the mean. Note that including the northbound vehicles on SR 436 in the travel time calculation increased the number of travel time matches by almost 50 percent. This increase in the sample size could result in more reliable travel time estimates or allow for a shorter averaging period with the same level of reliability.

It is interesting to note that the two travel time measurements, one for vehicles going straight through the SR 50 / SR 436 intersection and one for those turning left, provide a measure of the additional delay - about 36 seconds - for vehicles that turn left at that intersection relative to those that go straight.

In summary, the review of toll tag travel time operations for the SR 50 eastbound, from US 17/92 to SR 436, travel time segment identified the following observations regarding the use of toll tag readers for arterial travel times:

In the following sections, these factors will be examined on a system-wide basis.

4.5. Toll Tag Reader Operations

This section reviews the operation of the toll tag reader system, with the primary objective being to assess the reliability of the iFlorida toll tag reader network. The following list describes the primary measures were used to assess the level of service provided by the iFlorida toll tag readers. (Appendix A describes these measures in more detail.)

These measures were assessed for each iFlorida toll tag reader. Figure 45 summarizes this level of service information by presenting the weekly average for the level of service summed across all the iFlorida toll tag readers. Table 19 in Appendix A provides a more detailed listing of these measures for the toll tag readers on SR 50.

Graphic presents the weekly average for the level of service summed across all the iFlorida toll tag readers. Average reader level of service remained above 0.50 for much of the period from June 2005 through May 2007, with drops below 0.50 occurring in May 2006 and November/December 2006.
Figure 45. Weekly Average Level of Service for iFlorida Toll Tag Readers

Note that the deployment of the toll tag reader system was completed May 31, 2005, a time when the level of service was just above 50 percent. The main cause for the low level of service at this time was that the readers were actually deployed and tested from February through May 2005, but there was no regular monitoring of the status of the already tested equipment until August 2005. Many devices failed during this period. With so many devices failed, the maintenance contractor had neither the man-power nor the spare parts available to quickly make all the necessary repairs. The need for so many spare parts at the same time also overwhelmed the manufacturer, who was unable to promptly replace failed equipment.

It is important that a system to monitor and maintain field equipment be in place by the time any field equipment is deployed.

The challenge of maintaining the network was made more difficult because FDOT did not have access to a simple way to monitor the equipment. The contractor provided neither a tool nor instructions for monitoring the status of the toll tag readers. The CRS software, which displayed travel times produced by the readers, was having reliability problems of its own, preventing FDOT from using the lack of travel times in the CRS as an indicator of a failed toll tag reader. FDOT resorted to a manual process of reviewing the log files at the toll tag server to determine if a reader was producing toll tag reads and providing them to the server. This process, which initially required almost half a day to complete, is still in use today, though it has been refined and now only requires about one hour each day.5

By January 2005, FDOT had achieved a fairly high level of service for the readers. FDOT attributed the drop in the number of functioning readers that occurred at times during that year to failures due to lightening strikes, though network cuts also caused some failures. The high number of failures resulted in a long turn-around time in getting replacement parts, extending the downtime when a reader did fail. After noting the high number of lightening-induced failures, FDOT tested the use of power conditioners and new grounding methods to reduce these failures.

By January 2006, the level of service for the readers operated stably at about 90 percent for most of 2007 before the reliability dropped off again beginning in June 2007. When the warranty period for the readers ended in May 2007, FDOT did not receive adequate documentation or training on how to configure replacement parts. This delayed repairs, allowing failed units to accumulate while FDOT was conducting tests and developing repair procedures.

Reader failures will occur. The systems that rely on the readers must be designed to work around those failures.

Despite considerable effort and expense devoted to maintaining the toll tag reader network, the network averaged almost 10 percent failed readers when working well. At times, the failure rates were much higher. Because the CRS software did not include a graceful way of coping with missing travel time data resulting from these failures, such as filling in missing data with historical averages and alerting operators that historical averages were being used so that they can more closely monitor the affected segments, a failed reader usually resulted in missing travel information.

The reader level of service values provide an indication of how well each individual reader was operating. However, it requires reads fom two toll tag readers to generate a travel time estimate, so the level of service for a travel time segment should be a combination of the level of service values for the readers at the from and to nodes for that segment. The Segment Level of Service, LOSSegmentprovides a measure of whether the travel time system is producing as many travel time estimates as usual. (This measure is defined in Appendix A.) Figure 46 depicts the level of service for the iFlorida toll tag travel time links.

Chart depicts the level of service for the iFlorida toll tag travel time links. Data closely follows the level of service chart for the toll tag readers, except that the level of service is generally lowerdue to the level of service for the travel time segment being lowered whenever the level of service of the readers at either node of the segment drops.
Figure 46. Monthly Average Level of Service for iFlorida Toll Tag Travel Time Links

Note that it closely follows the level of service chart for the toll tag readers, except that the level of service is generally lower. This is because the level of service for the travel time segment is lowered whenever the level of service of the readers at either node of the segment drops.

4.6. Toll Tag Matching Efficiency

Another factor that impacts the effectiveness of toll tag readers for estimating travel times is the efficiency of the toll tag matching - what fraction of tag reads at the entrance to a travel time segment result in a travel time estimate when the vehicle exits the segment. The following list describes the main factors that affect this efficiency.

This section of the report explores the impact of these factors on the toll tag matching efficiency. Two measures were used for the toll tag matching efficiency. The entering efficiency was the fraction of toll tag reads for vehicles entering the travel time segment that generated travel time estimate. The exiting efficiency was the fraction of toll tag reads for vehicles exiting the segment that resulted in travel time segments. The average toll tag matching efficiency for each arterial segment is reported in Table 8.

Table 8. iFlorida Arterial Toll Tag Matching Efficiency
Road: Segment Length (miles) Lanes Maj. Intersections Tag Matching Efficiency
East/North West/South
Enter Exit Enter Exit
SR 50: SR 91 to SR 429 4.9 2 7 15 21 22 13
SR 50: SR 429 to SR 408 2.4 2 4 22 9 26 35
SR 50: SR 408 to SR 423 6.5 3 14 6 12 11 5
SR 50: SR 423 to SR 441 1.1 3 3 35 34 54 10
SR 50: SR 441 to I-4 0.8 2 3 48 26 12 45
SR 50: I-4 to US 17/92 1.0 2 4 29 43 23 48
SR 50: US 17/92 to SR 436 3.4 3 12 16 12 21 19
SR 50: SR 436 to SR 417 3.0 3 4 11 24 14 8
SR 50: SR 417 to SR 408 4.5 3 8 10 5 26 35
SR 414: SR 441 to I-4 4.3 3 2 33 14 12 5
SR 414: I-4 to US 17/92 2.0 2 7 25 42 16 38
SR 423: SR 417 to SR 528 2.8 3 3 25 49 32 27
SR 423: SR 528 to I-4 6.1 3 3 11 14 13 8
SR 423: I-4 to SR 408 3.0 3 3 19 11 11 18
SR 423: SR 408 to SR 50 0.7 3 3 14 54 47 34
SR 423: SR 50 to SR 441 3.2 2 2 16 7 15 14
SR 423: SR 441 to I-4 2.1 3 3 34 19 20 33
SR 423: I-4 to US 17/92 1.3 2 2 27 24 34 22
SR 436: SR 441 to SR 434 5.4 4 12 8 24 27 29
SR 436: SR 434 to I-4 1.8 4 7 50 20 25 30
SR 436: I-4 to US 17/92 3.1 4 12 12 12 18 14
SR 436: US 17/92 to SR 50 7.9 3 19 4 6 4 5
SR 436: SR 50 to SR 408 1.2 3 3 35 16 23 41
SR 436: SR 408 to SR 528 5.4 3 13 10 14 18 10
SR 441: US 192 to SR 417 4.7 3 7 14 12 16 22
SR 441: SR 417 to SR 528 4.4 3 8 2 8 8 16
SR 441: SR 528 to I-4 4.7 3 13 15 3 15 10
SR 441: I-4 to SR 408 2.2 2 5 15 21 29 17
SR 441: SR 408 to SR 50 1.2 2 7 39 31 53 21
SR 441: SR 50 to SR 423 3.4 2 4 2 4 13 31
SR 441: SR 423 to SR 414 3.4 2 4 4 2 29 28
SR 441: SR 414 to SR 436 3.5 2 5 26 56 16 39
SR 441: SR 436 to SR 429 2.2 2 4 54 15 57 57
US 17/92: SR 50 to SR 423 2.7 2 9 8 8 12 16
US 17/92: SR 423 to SR 414 2.3 3 4 23 36 18 14
US 17/92: SR 414 to SR 436 1.9 3 2 10 16 65 35
US 17/92: SR 436 to SR 417 8.8 3 15 5 3 2 2

In this table, the columns labeled East/North report the toll tag matching efficiency in the eastbound and northbound directions, while the columns labeled West/South are for the westbound and southbound directions. The Enter columns report the entering toll tag matching efficiency and the Exit columns report the exiting matching efficiency. Table 9 is an analogous table for limited access roads.

Table 9. iFlorida Limited Access Toll Tag Matching Efficiency
Road: Segment Length (miles) Lanes Maj. Intersections Tag Matching Efficiency
East/North West/South
Enter Exit Enter Exit
SR 91: US 192 to SR 522 4.2 2 0 78 93 52 63
SR 91: SR 522 to SR 528 6.9 2 0 37 32 19 60
SR 91: SR 528 to I-4 4.1 2 0 46 71 43 47
SR 91: I-4 to SR 408 6.1 2 0 84 27 78 55
SR 91: SR 408 to SR 50 7.0 2 1 17 49 29 25
SR 417: I-4 to Boundary 5.1 2 2 65 52 34 70
SR 417: Boundary to SR 434 6.1 2 2 33 48 52 32
SR 417: SR 434 to US 17/92 6.6 2 1 49 72 76 53
SR 417: US 17/92 to I-4 4.8 2 2 41 73 67 45

A number of observations can be made by examining the toll tag matching efficiencies listed in these tables.

The tag matching efficiency drops quickly as the segment length increases.

These observations are also demonstrated in Figure 47, with the trend line for limited access roads showing tag matching efficiency almost 40 percent higher than for arterials and the trend line for arterials dropping from about 40 percent for a 1-mile segment to less than 20 percent for a 5-mile segment.

Scatter graph features a trend line for limited access roads showing tag matching efficiency almost 40 percent higher than for arterials and the trend line for arterials dropping from about 40 percent for a 1-mile segment to less than 20 percent for a 5-mile segment.
Figure 47. Tag Matching Efficiency versus Segment Length

4.7. Latency and Travel Time Estimation

The travel times experienced by actual travelers are different from those measured by either loop detectors or toll tag readers.

During times when travel times are relatively stable, all of these values will be similar. When travel times are changing, such as when congestion is building, these times can be different.

Consider what would occur for a segment where the usual travel time was 10 minutes after a crash occurred that instantly changed the travel time experienced by travelers to 30 minutes by introducing a 20-minute delay for those vehicle passing through the crash location. With loop detectors,6 the observed speeds would quickly change from pre-crash to post-crash values, and the travel time estimated using loop detectors would change quickly from 10 minutes to 30 minutes.

For toll tag readers, the result would be different. During the first few minutes after the crash, vehicles already downstream of the crash location would continue to arrive at the end of the travel time segment and the system would continue to measure 10-minute travel times. After that, there would be a period during which few vehicles arrived at the end of the travel time segment because the vehicles were caught in the delay caused by the crash. The system might note that fewer vehicles were arriving than expected, but would have no travel time data to indicate that travel times had increased. After 30 minutes had passed, those vehicles delayed by the crash would arrive at the end of the travel time segment and the system would report 30 minute travel times.

Toll tag readers introduce a delay in producing travel time estimates equal to the estimated travel time.

Thus, travel times estimates produced by loop detectors respond quickly to changes in travel times that may occur. Travel time estimates produced by toll tag readers are delayed by the time required for the vehicles that enter the travel time segment to completely traverse the segment - the estimates are delayed by a time equal to the travel time estimate. While this delay is modest for short travel time segments, it can be long for longer travel time segments. For example, the iFlorida arterial toll tag reader deployment included a number of travel time segments that were 5 miles or more in length. Travel time along these segments during congested periods could exceed 30 minutes, which meant that the estimated travel times reflected travel times experienced by travelers entering the segment 30 minutes earlier rather than current traffic conditions. Thus, although toll tag readers produce accurate travel time estimates, in that they reflect travel times experienced by actual travelers, the delay in obtaining the estimate can limit the usefulness of the resulting estimates.

The latency inherent in toll tag based traffic monitoring systems could limit its effectiveness for incident detection.

This latency inherent in toll tag based traffic monitoring systems can also impact the usefulness of toll tag reader networks for incident detection. In order for automated incident detection to be useful, an incident should be identified within minutes of when it occurred. Because the measurement delay in a toll tag based traffic monitoring system is equal to the travel time for the segment, this type of system would provide effective incident detection only if the toll tag readers were spaced sufficiently close so that the expected travel time was less than time delay allowed for effective incident detection.

4.8. Summary and Conclusions

As part of the iFlorida Model Deployment, FDOT deployed a network of toll tag readers that were used to estimate travel times on seven Orlando arterials and on portions of several limited access toll roads. Soon after the deployment of these readers was complete in May 2005, FDOT discovered that many of them had failed between the time they were deployed and FDOT began monitoring them. From that time until early in 2006, FDOT struggled to achieve reliable operations with these readers. The reader network operated reliably for the first six months of 2006, though configuration problems with the CRS software made it uncertain whether the travel times being produced by this network were being used correctly to provide 511 travel time messages. Beginning in May 2006, the operational reliability of the toll tag network dropped again. Stable operations were restored in January 2007 and continued through June 2007, when the reliability dropped again as FDOT transitioned from the warranty period, where the deployment vendor was responsible for most repair activities, to a period where FDOT was responsible for maintenance. The CRS software also failed in June 2007, and it was not until November 2007 that FDOT had procured new software that would ingest travel times from the toll tag network and automatically populate 511 messages with those travel times.

Throughout this process, lessons learned regarding the operation of a toll tag network for estimating travel times were documented. These lessons learned are summarized in the following list, with more details provided in the preceeding sections of this report.




1 See the report Innovative Traffic Data Collection: Results of Field Test - Final Report prepared for FDOT on February 12, 2003 for more information.

2 See the report Field Test of the Potential for Using Probe Vehicles to Determine Arterial Travel Times in the Orlando Metropolitan Area - Final Report prepared for FDOT on July 1, 2003 for more information.

3 These figures include matches that resulted in very long travel time estimates, likely resulting from vehicles that made stops or other diversions between two toll tag readers used to make a travel time estimate. A production travel time calculation tool would filter out these outliers, resulting in lower percentage for matching toll tag reads.

4 This data is available at http://www.dot.state.fl.us/planning/statistics/trafficdata/ accessed March 10, 2009.

5 More information on the challenges faced in deploying the toll tag reader network is contained in the report titled iFlorida Model Deployment-Deployment Experience Study completed in May 2006.

6 The system response would depend on the location of the loop detectors relative to the congestion. If no loops were located in the congested area, then the system would not detect the additional delay. If loops were concentrated in the congested area, the system might over-estimate the delay. This statement assumes that loops were placed so as to accurately measure the crash-induced delay.

Previous | Next