Automotive technology’s potential use in rolling stock

[edwin_m]

Other than this I could see an interesting future application of computer vision technology to recognise signal states and speed boards to provide an in cab view without fitting balises, ETCS was defined long before realtime computer vision became possible on a practical level.

Generally the driver needs to know about a stop signal well before it becomes visible, in order to stop in time (as does any automatic train operation system). So you need some sort of communication of the signal aspects to the train, and given that it's probably easier to use the same thing to communicate the signals that are visible. Such a system (ERTMS level 2) also creates the prospect of saving a bit of money by getting rid of the signals entirely.

I was on the committee (as a lowly minute-taker) that eventually came up with TPWS as a SPAD mitigation measure. In about 1994 it was considering a wide range of options and machine vision came up as a way of enforcing red signals without having to fit a lot of extra trackside kit. I think the general view was that the technology wasn't sufficiently reliable, and a signal that was confusing enough to be a high SPAD risk would confuse the machine vision as well.

 

[squizzler]

The whole philosophy of rail travel is of guidance by mechanical means, not by picking your way along the road on sight. It made sense to augment then do away with the weaknesses of human sight for the stop and go aspects when the technology became available, hence the development of various protection systems and then in-cab signalling. Feel gives more certainty than sight, and railway technology leverages this.

The superiority of underlying principles of rail transport is conclusively demonstrated by the attempts to automate motorcars. Whilst a Waymo test driver struggles to bump and scrape its way with a couple of passengers along some well mapped streets in clement California, a high speed train can shift 500 people at 350kph in perfect safety using, I suspect, an order of magnitude less processing power to do so and an absence of vulnerable cameras and LIDAR scanners (obstacle detection can be delegated to the right of way in high risk areas like under highway bridges).