Want to know the real reason SANRAL wants people to sign up for eTags, and charges more for those who don’t? Because its cameras
suck aren’t reliable.
I’ve been chatting to people for the last week and gathering lots of confirmed reports that the eToll billing system is prone to inaccuracies. And from what I’ve seen so far, the inaccuracies only go one way: those who have been using the eToll roads without eTags have not been billed for all the gantries they have passed through.
People opposed to eTolls might find this to be good news. Not only is the system they’re opposing not working properly, but their accumulated eToll charges will also be less than anticipated. Perhaps it’ll even be grounds for another legal challenge next year. Certainly it makes a bit of a mockery of the cost and obnoxious appearance of the purple-lit gantries: one staffer in the htxt.africa office has an account on the SANRAL website, though not an eTag, and confirms that his account only shows seven charges, despite passing under at least ten toll gantries since the system went live on the 3rd of December.
And as I say, I’ve spoken to others who also believe that their cars have passed through the gantries unbilled.
These inconsistencies can not have been unexpected, at least not to SANRAL. Kapsch, the company that provides the equipment and software for the e-Toll, doesn’t publish figures for the accuracy of its optical character recognition (OCR) system. This is system that “reads” number plates and turns the image of XYZ123-GP into text that can be read. The most accurate OCR is said to be 98% effective, in ideal conditions. The highways of Gauteng are hardly ideal conditions, though. Cars travelling at different speeds; number plates displayed at different heights and angles; various types of number plates; dirt and grime; weather; water spray from the rain; lighting conditions – these are all things that affect the accuracy of an image recognition system. The two ANPR cameras listed on the Kapsch website might be high-quality items, but image sensors of just 1.3-megapixels and 1.4-megapixels show up their limitations.
All of this adds up to give a real-world accuracy, for ANPR systems, of around 90 – 94%. Another study at the University of Hertfordshire, in the UK, did in-depth analysis on the accuracy of ANPR system and found that the system often misreads certain letters, and also misreads letters that are in certain positions on the number plate. Failure rates – where a letter is misread – were as high as 34%.
There are ways around this. We know that SANRAL’s eToll gantries take multiple photos, from multiple angles. If the first shot fails the second could still be a success. And if there’s an invalid hit – and this is just speculation, based on how we guess such a system should work – the system could flag it up for manual review.
That right there – a manual review – could be one reason that fees for non-tagged cars are higher. Presumably, if such a system is in place, SANRAL would have to pay somebody to manually review those invalid hits. Then there’s the processing of those non-tagged accounts, sending bills, collecting payments, and so on.
It could also be argued that SANRAL is charging non-tagged users more because its cameras won’t pick up every time a non-tagged car drives under a gantry. You might get missed by this gantry, but you’re still paying double the price a tag user pays, so it kind of balances out. Basically, the fees could be higher because even with 90% accuracy, getting twice the money is better than getting half the money at 100% accuracy.
Finally, it all points to why eTags are being pushed so heavily. Yes, in a world where people completely agree to the eToll system and unconditionally trust the government, the eTag system would be fantastic. Convenient, easy to pay, and fast.
But right now it’s the only way SANRAL can guarantee a 100% hit rate for cars that drive through the toll gates. The cameras aren’t accurate, and the eTag never fails (or at least, we haven’t proved that it does yet).