E-toll problems on Carquinez bridge CA

E-problems on Carquinez bridge CA

Electronic tolling is getting a bad name in the Bay area. Local media have picked up the story of system integrator MFS having problems in tests on the Carquinez bridge with the first electronic toll installation for the California Department of Transportation (Caltrans). A typical piece in the San Francisco Chronicle: "Bay Area drivers have seen the electronic future -- and it's been sent back for more testing. That is the word now that an automatic toll collection device has flunked a three-month-long test on the Carquinez Bridge and gone back to the contractor for modification. The electronic device, technically called a transponder ..."

Contrary to that report the electronic transponder (e-tag) system passed all the tests at Carquinez meeting Caltrans 99.99% accuracy requirement. Hardly surprising. It is the same Texas Instruments e-tag and reader system that has been in use on three toll roads in southern California and elsewhere and performing without any significant problems.

But MFS's installation on Carquinez — the bridge which carries I-80 the main expressway between San Francisco and the state capital in Sacramento is to be the first of nine state toll bridges to be automated — involved not only e-tolling, but automatic vehicle classification (AVC) and a violation detection system (VDS) that does license plate recognition. The local rag named the wrong villain! It was these associated systems that failed recent tests, not the e-toll "device."

The VDS in September achieved an overall readability of violators license tags of 97.4% compared to 99.0% required, and the AVC performed at 99.1% accuracy compared to the contract requirement of 99.95%. Officials say the violation detection system only requires "fine-tuning" to get at, or very near, the target. However there is less certainty the company can meet the accuracy target in its vehicle classification. A table summarizing test results and expectations that we obtained shows the projected accuracy after refinements of the classification system and planned tests in March 1997 at 99.80% compared to the contract standard of 99.95%.

Video records were kept of the systems in operation to make the accuracy assessment. The serious failures in the fall tests are quite striking in the table produced to summarize results. 2-axle trucks had a classification accuracy of only 66.9%. Motor cycles were only correctly classified 62.5% of the time, and commuter buses and 2-axle buses (two different Caltrans classes) 66.7%. Light vehicle "3-axle combos" (cars with a trailer) were misclassified one time in ten. The positive way of looking at all this is to say that the inaccuracy is limited to classifying just a few vehicle configurations and that the system is performing pretty well on the majority of configurations. 96% of total vehicles were classified with 99.77% accuracy. 85% of the errors occured among configurations representing about 5% of the traffic flow. The subsystems had trouble distinguishing 'dually' (paired wheel) vehicles from normal single wheeled, and failed to distinguish goosenecked trailers and certain motorbikes, buses and motor-homes.

The classification system Caltrans has accumulated over the years is very complex according to Charles Price, head of the state agency's toll automation effort. Ideally, he told us, they would have simplified and rationalized the vehicle classes before automation. This is the approach favored by the staff of the New York State Thruway Authority for its 1025km 61 plaza system. The Thruway is quickly and successfully implementing electronic toll collection for passenger cars but it has deferred automatic vehicle classification as too difficult given what one official there described as a complex and archaic set of vehicle classes much like California's. Caltrans' Charles Price told us the state made the decision that it would be too controversial politically if vehicle classes were altered to make electronic tolling easier, so it asked for proposals from system integrators in 1993. Several companies that would normally bid for a project of this value considered the Caltrans bridge job impossible, and MFS was the sole bidder for the $30m contract.

To overcome the problem of distinguishing dual tires from single tires it is normal to install an angled treadle (placed at 45 degrees to the traffic flow) that remains depressed longer by a dual tire than by a single tire but this is not acceptable on Caltrans toll plazas which are located on bridge decking. It might damage the strength of the road deck to jack-hammer new slots. MFS is developing a totally new treadle to fit a regular treadle slot — with a special segmented strip that simply measures the width of vehicle tires according to the number of separate small segments that get pressure. And to better pick up trailed vehicles with gooseneck hitches it is installing an unconventionally tall light curtain, and also a canopy-mounted vehicle profiler. And it has to write new computer code to process all the new data, and get the stuff working together. Then it will be retested in March next.

One benefit from such a fiasco might be a serious discussion of accuracy standards. MFS's Bill Brogdon says there has been an unhealthy "code of silence" about the accuracy of automated systems and the costs. How much accuracy is enough? A starting point for tollsters is the accuracy of human toll attendants. Many studies have been done by toll authorities apparently, but the results of most kept in-house, because any reports of inaccuracy have the potential to attract criticism and may also upset employees. Anecdotally it is said that these studies have shown human accuracy ranges between about 85% and 97%. It is unclear why machines need to be specified for several orders of magnitude more accuracy than humans. There are usually sharply diminishing returns on any system striving for perfection, since the last few tenths of a point accuracy provide little benefit at great cost.

In Caltrans' case there is a special reason why the automatic vehicle classification system is important. Most tollsters encode vehicle classification data in the transponders or e-tags, so that vehicle classification equipment acts as simply a check or backup to the encoded information rather than as a primary system. Caltrans by contrast has decided that such coding of the e-tag would "invade privacy" so its vehicle classification equipment is a primary system. (Contact Charles Price Caltrans 510 286 4478, Bill Brogdon MFS 402 233 7729)