A full barrier crossing scans itself with radar before it lets a train through, and an obstruction holds the protecting signal at red until it is cleared. Ours reports an obstruction of about a metre and a half, in the same place, most nights, with no train due and the barriers up.
I am a signalling technician and a level crossing on a rural line is on my patch.
How the obstacle detection works.
When the barriers lower, a radar scanner at each corner sweeps the road area between them. If anything is detected the protecting signals stay at danger and an alarm goes to the signaller, who has cameras and can decide what to do.
The radar reports the size and position of what it sees. It is tuned to ignore a pigeon and see a pram. It is tested weekly with a target of a defined size placed on the crossing, and the test is logged.
Outside a barrier sequence the radar is idle. It has no reason to be looking.
At about three in the morning, most nights, the system logs an obstacle detection of about one and a half metres by half a metre, in the four foot of the up line, near the centre of the crossing.
The barriers are up. There is no train in the section and none due for two hours. The scanners should not be running and the log says they were not commanded to.
The detection has a duration. It is usually between forty seconds and two minutes, and then it clears, and the log records it clearing.
There is no signaller action, because there is no signal to hold and nothing is approaching. The alarm shows at the signalling centre and the signaller looks at the cameras and sees an empty crossing, and logs it as a false detection, and I get an email in the morning.
What has been checked.
The scanners. All four replaced in 2022 in a planned renewal and re-commissioned with the target tests.
The control logic. The obstacle detection is not supposed to run outside a barrier sequence, and the supplier has confirmed from the logs that it does not. The detection events are timestamped and arrive as detection events, and their software does not have a state in which an idle scanner reports a target.
Their engineer wrote that the scanner is being triggered by something and that he would like to know what.
The cameras. Two, both recording continuously, both reviewed on eleven occasions for the minutes concerned.
The crossing is empty. The road is empty, the barriers are up, the lights are off, and the picture does not change.
What I tried in 2023.
A radar sees things a camera does not, which is why it is a radar and not a camera. So I took a thermal imager and a colleague and we sat in the van by the crossing on three nights.
The third night it did it, at eight minutes past three. The alarm came through on my phone from the signaller and I was already looking at the crossing.
There was nothing on it in visible light or on the thermal.
The radar log for that event shows an object of one and a half metres by half a metre, in the four foot, that moved from the up side of the crossing to the down side over about ninety seconds at walking pace, and then cleared at the edge of the scanned area.
That is the first time I had looked at the position channel over the course of an event rather than the summary, and it is why I stopped sitting in the van.
What is done about it.
It is a known false detection at that crossing, with a fault reference, and it costs the signaller thirty seconds a night to look at a camera and press a button.
The safety review concluded that a spurious detection outside a barrier sequence has no safety consequence, which is correct: a false obstruction does not clear a signal, it can only hold one, and nothing was being held.
I have asked for the scanners' detection sensitivity to be examined and was told, rightly, that a radar that reliably reports a pram sized object at the centre of a crossing is a radar that is working.
What I do.
I do the weekly target test and the maintenance and I read the emails.
The one change is the target test itself. You place a target on the crossing and confirm the system sees it, and the target is a metal cylinder about a metre and a half high on a stand, which is what a person stood on the crossing looks like to a radar.
I used to do that test at the start of the day, at about half six in winter, when it is dark.
I do it after nine now, in the light, and the reason is that in February 2023 I put the target on the crossing at twenty to seven, in the four foot of the up line, and walked back to the cabinet to confirm the detection.
The log had two targets on the crossing. Mine, and another of the same size, a metre and a half from it, that had been there for a minute before I arrived and cleared about ten seconds after I looked up.