[Go Back]

Comments on ITE Journal Article on GPS Accuracy

There's an nice article in the Feb. 2003 ITE Journal on the accuracy of GPS data for field traffic studies. A Tru-Traffic user asked for my comments, which I'm posting here in case others are interested in those as well:

I like their suggestion for interpolation to reconstruct missing data. Right now, Tru-Traffic will stop recording a trip log if the data stream is interrupted for more than 5 - 6 seconds, by default (the actual time is a user preference). The assumption is that something bad has happened in that case. I'm considering using interpolation in cases where the interruption is long, but not too long. User feedback will be very valuable in making this decision. Two users have reported an error message

Stopped recording trip log.
The data stream from the GPS receiver was interrupted.
Make sure the data cable is securely connected to your computer and GPS receiver and the power to the GPS receiver is reliable.

If this is persistent or a general problem, I might be able to have Tru-Traffic recover from the signal interruption and do something with interpolation to reconstruct the missing data where practical. But this partly depends on the circumstances under which the error occurs. If you're getting this error, please contact me so I can send you a diagnostic version of the software to help us track learn more about the problem.

A question that some people raise is how does the inaccuracy of the GPS affect the travel time and delay report in Tru-Traffic.

At first blush, an estimated imprecision of 10 - 15 m (33 - 50 ft) sounds like it might be pretty significant for the TT&D report. For example, the intersection-to-intersection travel distance calculation is made by connecting the dots between numerous GPS readings and summing the straight-line distances. There are typically 10's or even 100's of readings between neighboring intersections in this calculation, and if each reading can be off by up to 10 - 15 m, then that ought to seriously degrade the calculation, right? Nevertheless, users compare the distances calculated from various recorded trip logs and find that they often agree to 1 m or less. How can this be?

Although not immediately obvious, one of the figures in the ITE Journal article illustrates why even the least accurate GPS unit tested still gives pretty good and consistent results in the TT&D report. Figure 1(c) shows the deviations in measured coordinates over a continuous 12-minute period. For the least accurate GPS, the deviation starts off at 10 m (33 ft) and drifts to 15 m (49 ft) by the end of the recording period. During the entire 12-minute period, the deviation is never less than 10 m. But (and here's the key point) the deviation drifts at a fairly slow rate -- at its worst, the deviation drifts at the rate of only about 1 meter per minute.

During the time spent traveling between intersections (including the time waiting in the queue), typically a few minutes or less, the drift in the deviation in the worst case is only a few meters. Although the deviation may be fairly large, it doesn't change much during that time. Because of the typical size of intersections and the way Tru-Traffic does its calculation, the calculated travel distance is usually insensitive to the absolute deviation. It is sensitive to the drifts in the deviation during that travel time, but that number is fairly small. Between trip logs (separated by minutes, days, or months), there might be substantial drift, but during the actual travel time between two neighboring intersections, the drift is pretty insubstantial. Consequently, between various trip logs, Tru-Traffic might end up measuring the travel distance using different reference points at the intersections. During one run, it might effectively measure the travel distance from center to center. During the next run, it might effectively measure the travel distance between the NB stop bars, and during the next run, it might be between the SB stop bars (on a NB run). But within a single trip, it uses pretty much the same reference points at both intersections (as a result of the slow drift), so the calculated distance isn't affected too much by that drift.

This is why the TT&D report gives consistent results even for relatively imprecise GPS units.

By the way, there's an error on page 34 of the article. The article mentions default messages from the National Electrical Manufacturers Association (NEMA). In fact, the messages are from the National Marine Electronics Association (NMEA) www.nmea.org.

Last revised: March 08, 2004