Signal Loss
The video stream timeout has been exceeded. Power cuts, camera overheating, a failed switch, a cut cable: the camera is simply gone, and this is the alarm that says so first.
Camera Tamper Detection and Video Quality Control
The worst camera fault is the one nobody notices. A lens sprayed on Tuesday is discovered three weeks later, when an investigator opens the archive and finds a wall. TamperTrack watches the estate itself and raises a fault the moment it happens, as an event, on the same monitor as everything else.
The Detector
TamperTrack Watches the Cameras, Not the Scene
Every other module on this site detects something in front of the camera. TamperTrack detects something wrong with the camera: no signal, too few frames, a stream suddenly heavier than it should be, a view that has moved, a lens that is covered, dirty, out of focus, blinded by light or lost in the dark. Each condition is its own alarm with its own priority and its own sensitivity, switched on per camera on the Detectors tab, and it runs beside whatever analytics module the channel already carries.
The current detector is a neural network trained on a proprietary dataset, introduced in 2025 to replace a simpler comparison. It identifies full and partial occlusion, whether the cause is a hand, a spray can or a drone, while ignoring the natural changes an estate produces every day: lighting shifts, snowfall, a shadow moving across the scene. Repositioning is judged against a stored reference image, and the platform can regenerate that image automatically while a camera is being settled in difficult light.
The point is the estate, not the camera. On a network of thousands of cameras most are only ever viewed after something has happened, so a camera that has quietly stopped seeing is a gap that stays open until the day it matters. TamperTrack turns that gap into an event with a priority, an alarm monitor and a search filter, so the operations team fixes it on Tuesday rather than the investigator finding it three weeks later.
Nine Conditions
Nine Alarms, and Why Each One Fires
Every alarm below is switched on separately, given its own priority, and tuned with its own sensitivity slider on the camera it protects. The causes are the ones IREX has documented, not a guess.
Low Frame Rate
Frames arrive below the minimum you set against the camera’s normal rate. It usually means a misconfigured camera or a link without the bandwidth, and it is the fault that makes every other detector on the channel worse.
High Bitrate
The average bitrate has spontaneously risen by more than a fifth. A camera flooding the network is an infrastructure problem before it is a video problem, and this alarm exists so an operator can step in before the platform is overloaded.
Camera Shifted
The view no longer matches the reference image. Vandalism, weather, a construction crew, a loose bracket: the camera may still be perfect and now watching the wrong thing, which is the failure a human operator never notices.
View Obstructed
A significant part of the frame is blocked, by paint, a bag, a hand, tape or a drone. You set the share of the region of interest that has to be covered before the alarm fires. The guide calls it out as a possible sign of planned wrongdoing.
Image Blurred
The whole picture has gone soft: wet or muddy spots on the lens, condensation inside the housing, heavy fog. Distinct from defocus since release 4.27, so maintenance knows whether to clean or to service.
Defocus
The optics are out of focus, typically an autofocus error. A separate event type from a dirty lens for the same reason: the fix is different.
Scene Too Bright
Overexposure from an unwanted reflection, a sensor or aperture-controller fault, or the sun sitting where it did not sit when the camera was installed.
Scene Too Dark
Underexposure from a failed infrared illuminator, an aperture controller stuck shut, vandalism, or the light simply not being there. Tuned per camera, because a lobby and a coastal mast do not go dark the same way.
On Camera
The Fault Is the Picture
A Camera Pointing at a Wall
Nothing is wrong with this picture. It is sharp, correctly exposed and entirely useless, because the camera has been knocked upward and rolled on its bracket and now covers a blank wall and the sky, with a sliver of the street it was installed to watch along the bottom edge. This is the fault a human operator never notices, since the feed is still live and looks healthy in a thumbnail, and it is the one a comparison against the reference image finds immediately.
The Reference Image It Compares Against
Camera shifted is a comparison, and this is what it compares against: a reference frame stored for the camera in the detector’s own settings, here a parking garage bay under a working floodlight. The module continuously checks the live view against it and alarms when the view has moved. In low light the comparison can over-fire, so the guide describes switching on automatic regeneration of the reference image until the camera settles, then switching it off again once the false-alarm rate is acceptable.
Tuning
Three Controls per Alarm
There is no fleet-wide setting. Each camera carries its own alarm set, because a station platform and a rooftop mast fail differently and should be judged differently.
- 01
Pick the Alarms and Their Priority
Tick the conditions that matter on this camera and give each a priority rating. Signal loss on a perimeter camera might be Critical; scene too bright on a lobby camera that catches the morning sun might be Low, or switched off.
- 02
Set the Threshold
Each alarm has a sensitivity slider, and two carry a number as well: the minimum frame rate below which Low frame rate fires, and the share of the region of interest that has to be covered before View obstructed does. The guide is candid that the right values are found experimentally, camera by camera, between false positives and false negatives.
- 03
Settle the Reference Image
Camera shifted is judged against a stored reference frame. In poor light that can produce too many alarms, so Autogenerate reference image updates the reference whenever the camera shifts until a steady state is reached, and is then switched off so that a real repositioning is caught again.
Where It Surfaces
Three Places a Fault Shows Up
- The Alarm Monitor
- A camera fault is an event like any other, scoped by camera and type onto the monitor of the role that owns it. Release 4.13 added the high-bitrate alarm precisely so it could be routed to a dedicated monitor and acted on before the platform is overloaded.
- The Camera Status
- On the Cameras page and in Settings, an affected camera carries a Video issues status. Hovering shows the issue type, such as Scene too bright or Low frame rate, and the status filter searches by issue type, so a maintenance team pulls the list of dark cameras without opening a single feed.
- A Question in Plain Language
- “Show me cameras in Zona 1 with video quality issues” is a working Ask IREX query on our own home page. The answer is the set of frames that are visibly, diagnosably broken, with the fault named on each.
Scope
What It Does and Does Not Do
- It detects and alerts. It does not raise maintenance work tickets or drive a ticketing system by itself; events reach external systems through the alarm monitor and the API like every other event class, and the ticket is yours to raise.
- It runs as a detector alongside whatever analytics module the camera already carries, so watching camera health costs nothing in module capacity. It can also run alone, with only View obstructed enabled, on cameras that carry no other analytics.
- Since release 4.27 a dirty lens and a defocused lens are distinct events, so the maintenance response can be precise: clean, service, or re-aim.
- IREX publishes no detection or false-alarm figure for it. Thresholds are set per camera during the pilot, on your own estate, in your own light.
Where It Runs
The Estates Where a Blind Camera Costs Most
Every estate needs its cameras working. These are the ones where nobody is standing at most of them, so the alarm is the only way anyone finds out.
Real-Time Crime Centers
A blinded camera is not a gap somebody discovers three weeks later during an investigation. It arrives on the wall as an event.
Read MoreCritical Infrastructure
On a perimeter measured in kilometers, a camera that has stopped seeing is an unmonitored kilometer.
Read MoreSmart Cities & Municipalities
Thousands of cameras across a city, most of them never watched live. Camera health is how the estate stays worth having.
Read MoreBus Shelters & Transit Stops
On an estate nobody visits, an obstructed or repositioned camera is an unmonitored site until the alarm says otherwise.
Read MoreRetail
A sprayed, knocked or defocused camera raising its own alert at every site is how a few hundred stores stay covered between service visits.
Read MoreParking Facilities
On an unlit deck a blind camera can otherwise go unnoticed for weeks.
Read MoreFAQ
Does it create a maintenance ticket?
No, and we would rather say so than let a specification assume it. TamperTrack raises events: a fault becomes an alarm with a priority on the monitor of the role that owns the camera, and a Video issues status on the camera itself. Events reach external systems through the alarm monitor and the platform API like every other event class, so a ticket in your own maintenance system is an integration scoped during design, not a feature of the module.
Will it fire every time a cloud passes or it snows?
That is what the 2025 detector was built not to do. It is a neural network trained on a proprietary dataset to identify full and partial occlusion while ignoring natural, non-threatening changes such as lighting shifts and snowfall. Each alarm also has its own sensitivity slider, tuned per camera during the pilot, and the reference image behind Camera shifted can be regenerated automatically while a camera settles in difficult light.
Does it need its own analytics capacity?
No. It runs as a detector alongside whatever module the camera already carries, and it can run alone on cameras that carry no other analytics. The guide publishes no computational rating for it, so we do not quote one.
Can it tell vandalism from weather?
It tells you what the camera is experiencing, and the causes overlap: Camera shifted covers a struck bracket and a storm alike, Image blurred covers mud on the lens and heavy fog. The distinct event types since release 4.27, a dirty lens and a defocused lens as separate alarms, exist so that the maintenance response can be right even when the cause has to be confirmed on site. What the module does not do is name the culprit.
What happens to the other detectors on a faulty camera?
They keep running on whatever frames arrive, which is exactly why the fault matters: a dark, blurred or shifted camera produces worse detections from every module on it. TamperTrack is how the platform tells you that the input has degraded rather than letting the analytics quietly degrade with it.
Does this run on its own or does someone have to watch it?
It runs on its own and alerts a person. Detections are signals for a human to verify: no response runs autonomously, and the verification and decision are logged against the same Case ID as the detection.
Count the Cameras You Are Not Watching
On most estates that is nearly all of them. A pilot puts a health alarm on each one and tells you, on the first morning, how many were already blind.