Plain Line Pattern Recognition (PLPR)

[jacko2327]

Hi, was wondering if anybody could give me some detailed info on what the purpose of the PLPR is, why it is used and what it exactly does. I have managed to find bits of information about this but nothing sufficient.

Thank you in advance

 

[swt_passenger]

I'm pretty sure it's a way of using track recording cameras automatically to find defects, rather than someone having to study the video frame by frame, and I think the basic technique is that successive video recordings are compared by a computerised system that flags up differences between current and previous recordings.
Google found this:

Plain line pattern recognition system for Network Rail

Rail Vision developed InfraVision RailVis software for Network Rail that takes as input a sequence of top or angular view images and generates both visual and statistical outputs. The software operates on a sequence of images taken by line or area scan camera with NMT train speed up to 200 km/hr. The hardware system was originally installed by Cybernetix of France.

InfraVision RailVis software provides the image analysis software to process these images for detecting rail features, defects and condition that can be efficiently fed into asset and track maintenance systems. The software at present is capable of automated detection of the following components using a range of camera views through image analysis: sleepers, check rails, switches, level crossings, clamps, adjustment switches, clips, defects, fishplates, tpws, and joins. In addition, the software can detect missing components (clips, housings, etc.) and classify them. The front end software presents the results of analysis and reports in a format required by Network Rail.

http://www.rail-vision.co.uk/projects

 

[Hydro]

I'm pretty sure it's a way of using track recording cameras automatically to find defects, rather than someone having to study the video frame by frame, and I think the basic technique is that successive video recordings are compared by a computerised system that flags up differences between current and previous recordings.
Google found this:

Plain line pattern recognition system for Network Rail

Rail Vision developed InfraVision RailVis software for Network Rail that takes as input a sequence of top or angular view images and generates both visual and statistical outputs. The software operates on a sequence of images taken by line or area scan camera with NMT train speed up to 200 km/hr. The hardware system was originally installed by Cybernetix of France.

InfraVision RailVis software provides the image analysis software to process these images for detecting rail features, defects and condition that can be efficiently fed into asset and track maintenance systems. The software at present is capable of automated detection of the following components using a range of camera views through image analysis: sleepers, check rails, switches, level crossings, clamps, adjustment switches, clips, defects, fishplates, tpws, and joins. In addition, the software can detect missing components (clips, housings, etc.) and classify them. The front end software presents the results of analysis and reports in a format required by Network Rail.

http://www.rail-vision.co.uk/projects

That's the old obsolete Cybernetix system, which never really worked properly.

The current version, developed by Omnicom Engineering, uses a bank of cameras and lasers to produce HD laser scanned images of the track in several different planes. These are then processed via machine vision software using pattern recognition algorithms to identify broken, missing, misaligned, or damaged track components.

It's very much in the early stages, it's fitted to an MPV and the NMT's Development Vehicle. I've spoken to one of the guys developing it and it seems to be coming along very well, with a few hiccups.

It'll be trialled on the WCML soon. The MPV will be out at some point at night on WCML South doing all the S&C. It was envisaged that a fleet of DVT's will carry this, coupled to the new 57's, but that seems to have gone quiet and rumours are it'll go on existing test coaches.

EDIT - should mention the MPV was an S&C inspection vehicle using different style of pattern recognition kit I recall, but it's the same theory. The PLPR stuff, I think, uses different algorithms for the different components in plain line.

 

[The Informer]

That's the old obsolete Cybernetix system, which never really worked properly.

The current version, developed by Omnicom Engineering, uses a bank of cameras and lasers to produce HD laser scanned images of the track in several different planes. These are then processed via machine vision software using pattern recognition algorithms to identify broken, missing, misaligned, or damaged track components.

It's very much in the early stages, it's fitted to an MPV and the NMT's Development Vehicle. I've spoken to one of the guys developing it and it seems to be coming along very well, with a few hiccups.

It'll be trialled on the WCML soon. The MPV will be out at some point at night on WCML South doing all the S&C. It was envisaged that a fleet of DVT's will carry this, coupled to the new 57's, but that seems to have gone quiet and rumours are it'll go on existing test coaches.

EDIT - should mention the MPV was an S&C inspection vehicle using different style of pattern recognition kit I recall, but it's the same theory. The PLPR stuff, I think, uses different algorithms for the different components in plain line.

We use the Omnicom recordings for quick reference when planning work. Its like cab riding in high definition. I have the whole Wales and Marches route on a 1TB hard disk! It really is a handy resource.

 

[Hydro]

We use the Omnicom recordings for quick reference when planning work. Its like cab riding in high definition. I have the whole Wales and Marches route on a 1TB hard disk! It really is a handy resource.

The full Omnicom survey, the OS3D stuff, is incredibly comprehensive and accurate. Their kit is mounted on the Track Recording Unit and we go out with them when they do the survey. Every time the direction is changed, the kit needs 5 minutes to get a ridiculously accurate fix on the location and bearings of the whole train, during which even walking around can disturb the whole process. I've been out with them doing the NW area a fair bit recently.

 

[Trog]

The full Omnicom survey, the OS3D stuff, is incredibly comprehensive and accurate..... Every time the direction is changed, the kit needs 5 minutes to get a ridiculously accurate fix on the location and bearings of the whole train, during which even walking around can disturb the whole process.....

But do they then use the data?

As old Omnicom is often 20 odd yards out on the mileage, and the new HD stuff is often over 100 yards out. Whenever I use it I have to calibrate it against two or three mileposts and work out what to add or take from the mileage it tells me. Its useful but you have to know what you are doing, take care and know its limitations.

 

[Hydro]

It depends on the quality of the positioning data that updates the system after the initial set off I think. My understanding of it is it's a GPS feed, plus 2 tacho inputs averaged against each other, plus gyro units under the train that all get mashed together to create the trains position right down the to the line. This forms the basis of the Real Time Positioning System, found on all test trains. Now RTPS has a few quirks in itself, i.e. it can lag, it can lose the GPS fix, it hangs up on junctions, and it likes to decide it's on the wrong line if the gyros have a wobbly. The tacho inputs are generally quite good, except if the units been for tyre turning and no-ones updated the feed count to reflect the smaller wheel diameter. A shortcoming in any of these can lead to the system being out of sync with the mileages displayed.

It's the future apparently, it's to control the geometry recording on new systems, rather than us manually synchronising the system to mileposts. But then, we can correct the synchronisation ourselves if somethings not right, unlike the RTPS system.

Update on PLPR, apparently we are not getting dedicated PLPR trains any more, rather it's being fitted to existing test coaches, and the testing frequency is being upped dramatically to fit in with inspection timescales. DVT's are still on the way, and apparently "they" (whoever they are) think that 3 x 57's will be available for standard monitoring work. Expect to see test trains nearly 24/7 around the network in the future.

 

[jacko2327]

Thanks for your responses they were extremely helpful.

Thanks

 

[John Webb]

....... The tacho inputs are generally quite good, except if the units been for tyre turning and no-ones updated the feed count to reflect the smaller wheel diameter........

The 'old' way of measuring distance, for instance in dynamometer cars, was a 'ninth wheel' - an independent wheel running on one of the rail surfaces but not used for supporting the vehicle. Such a system was used to measure Mallard's 126mph in 1938. Is there any reason why this system couldn't be used today?

 

[Hydro]

I don't see why not, in theory. A tacho off the standard wheels does keep things rather simple, well...as simple as the set ups can be under these trains.

 

[Ploughman]

It's the future apparently, it's to control the geometry recording on new systems, rather than us manually synchronising the system to mileposts. But then, we can correct the synchronisation ourselves if somethings not right, unlike the RTPS system.

After wheeling out numerous renewal sites and recording milepost locations I wouldn't trust their accuracy one bit.
The amount of them that I found to be within 5 yds of their supposed location was very few. Some being upto 90 yds out of place. When comparing to old records and diagrams some could be explained by being placed as near as possible to the correct position but having an MP in the middle of a Level crossing would not go down well with car drivers. These are usually listed on the diagram as being so many yards out of place but the rest is, to me, down to trackside work in the past removing the post and then guessing that it should be about here somewhere if they ever got round to replacing it later.

 

[Hydro]

Every mile is verified for accuracy within the in board systems as the programming stems from the GEOGIS/infrastructure registers. If the mileposts aren't accurate, it actually compensates for it using something called an Imperfect Milepost Statement (IMPS). This gives a percentage that changes depending on whether it's short or long (i.e. 100% being a perfect mile with everything in the right place - lower or higher will indicate how reliable that distance is). We don't synchronise on poor IMPS unless theres no other way.

It ain't perfect, so that's where judgement, area knowledge and skill come in - to average out any drift and keep the systems as accurate as possible.

 

[Trog]

I know of one milepost that rotted off at ground level and fell over.
The top part was then fed into a nearby latticework OHL mast, where it has stood ever since. As that was over twenty years ago I suspect I am the only person who remembers that the milepost is out and why.