Camera-based production monitoring: what an iPhone actually sees
Most of the production monitoring sold today asks you to wire your PLC into a gateway, push the data to a backend, and then build dashboards on top. That pipeline has been the default for 15 years. It works when the PLC is modern and the integration team is patient. It does not work when you need shift visibility on a line by Friday.
Camera-based production monitoring is the alternative path. A refurbished iPhone, a mount, a ring light, and a vision model that has learned to read the line. The data flows the same direction. The signal source is different.
This piece is the visual version of the explanation. What the camera sees on a real line, what it cannot see, three live setups summarized in 60 seconds each, and the mounting basics that make the difference between a good feed and a useless one.
What the iPhone camera sees on a production line
The first thing the camera sees is the part itself. Every cycle, every part, every shift. The count that arrives in your dashboard is the count of parts the camera actually saw, not the count the controller said it produced. On most lines these two numbers differ by 1 to 4 percent over a week. Camera-based count is closer to what you ship.
The second is the cycle. The time between part 1 arriving at the outfeed and part 2 arriving at the outfeed is your real cycle time. The PLC tag often tells you the planned cycle time. The camera tells you the actual one. The drift between them is your performance loss.
The third is the stop. When the part flow stops for more than a configurable threshold (typically 8 to 15 seconds), the camera flags it. The operator picks one of the eight reason codes we cover in our downtime tracking software guide. The stop, its duration, and its reason all flow into the same dashboard.
The fourth is the changeover. The transition from "last good part of SKU A" to "first good part of SKU B" is one of the largest swings in OEE on most lines. The camera sees the disassembly, the tool change, the trial cycles, and the moment good parts start flowing again. The changeover duration becomes a separate metric in the dashboard instead of being lumped into "unplanned downtime."
The fifth is the rejected part. If the line includes an inspection step (we have written about that in production monitoring system options), the camera flags parts that fail the visual check before they leave the cell. The reject rate, the failure mode, and an image of each failed part flow to the dashboard. The operator does not have to remember which parts were bad.
[VIDEO PLACEHOLDER: 30-second clip of an iPhone watching a snack-packaging line outfeed, with overlay graphics showing the count incrementing in real time.]
What it cannot see (the real limits)
We have spent the last three years deploying camera-based monitoring on a wide range of lines. Some setups are clean and others are not. Here are the limits we have learned to name up front.
A camera cannot see inside a closed enclosure. If the line is a CNC machining center with coolant spray on the windows, the camera at the door sees the operator opening the door, not the cycle inside. PLC integration is the right answer for those cells.
A camera cannot read a controller register. If you need cycle time accurate to 10 milliseconds because you are tuning a servo motion profile, the camera is too coarse. The PLC has that data already.
A camera does not magically work in any lighting. Plant lighting is more inconsistent than most buyers expect. A ring light at the station fixes most cases for about 60 euros. A few cells (welding arcs, strong reflections off polished metal, extreme low light at the back of a warehouse) need a more careful setup or a different approach.
A camera does not solve a process problem. If your line runs at 65 percent OEE, the camera will show you the 35 percent loss in fine detail. The decisions about what to fix and in what order are still yours. The camera is a sharper instrument, not a different game.
[VIDEO PLACEHOLDER: 20-second clip showing a camera struggling in a difficult lighting condition, then the same setup after a ring light is added.]
Three live setups in 60 seconds each
Three deployment shapes that cover most of the lines we see in mid-sized plants. None of these name a customer; the point is the setup pattern.
Setup A: snack packaging line. Refurbished iPhone mounted 1.2 meters above the outfeed conveyor, ring light at 45 degrees. The camera counts bags, flags stops longer than 10 seconds, and watches for visible seal defects. Total hardware cost to get the line running: about 900 euros. Time from unboxing to first dashboard data: 47 minutes on the day we recorded the deployment video.
Setup B: ceramic tile inspection. Two iPhones, one at the kiln exit and one at the sorting station. The first counts tiles and reads the surface for cracks and color variation. The second confirms the sorter put each tile in the right pallet. Together they replace a manual inspection step that used to require an operator with a clipboard. Hardware total: 1,400 euros for the two stations.
Setup C: assembly cell with three SKU families. One iPhone watching the cell entrance, one watching the outfeed. The camera reads the SKU code on the work order at the entrance, switches to the right model for the current run, and counts and inspects parts at the outfeed. Changeover time is captured automatically as the gap between SKU A ending and SKU B starting. Hardware total: 1,100 euros.
[VIDEO PLACEHOLDER: three short clips, one per setup, 20 seconds each.]
The pattern across all three is the same. Hardware under 1,500 euros per line. Setup time of an afternoon. No PLC integration. Data flowing into the same dashboard regardless of the line shape.
Mounting and lighting basics
If you are going to try this on one of your lines, three mounting decisions matter more than the rest.
Distance from the cycle. Too close and the camera frame misses the full part. Too far and the resolution drops below what the vision model needs. For a part 10 to 30 cm in size, mounting distance is typically 0.8 to 1.5 meters. Test the framing at the planned mount location with a phone camera before drilling holes.
Angle to the part. A camera looking straight down sees only the top surface. A camera at 30 to 60 degrees from vertical sees the top and one side. For most counting and inspection use cases, a moderate angle gives better data. For barcode reading at the work order, straight-on is better.
Lighting consistency. Plant lighting changes through the day, especially near windows or skylights. A ring light or a small dedicated LED panel at the station gives the camera a consistent base level that does not drift. The cost is 40 to 80 euros per station. The data quality difference is large.
The mounting kit we ship with most deployments fits these three rules and goes up in under 30 minutes. For lines that need something custom, our team does a free 15-minute mount review on video call.
[VIDEO PLACEHOLDER: 25-second clip of a mount being installed, with the on-screen camera view updating as the angle is set.]
How this differs from security cameras and consumer photo gear
A note for readers who land here looking for security cameras, IP cameras, bullet cameras, or PTZ camera setups. This piece is not about that category. Process monitoring on production lines inside a manufacturing plant is a different problem from CCTV. The camera does not pan, tilt, or zoom across a wide area. It sits on a fixed mount above one cycle and reads the part. Pan-tilt-zoom feature sets and the wide-area event recording you would expect from a security camera are not what we use. The data we care about is a structured per-cycle record (count, cycle time, stop, reason), not a live streaming video archive.
The same distinction applies to consumer photo gear. We do not use a DSLR, mirrorless cameras, compact cameras, or medium format bodies for production monitoring. Autofocus, shutter speed in a creative sense, zebra stripes, or hot-shoe accessories are not relevant to the work. What we use is the iPhone camera that already lives in every pocket, on a fixed mount above the cycle, with a known field of view and predictable lighting. The CMOS sensor and global shutter behaviour matter for the vision model, but the way they matter is "is the part sharp every cycle," not the photographic vocabulary the gear-review sites use.
What we share with both categories is the basics. Video quality and image quality depend on lens, sensor, and lighting. Rugged housings matter in industrial environments where coolant, dust, and vibration are present. A solid mount (whether a magnetic clamp or a small tripod-style stand) keeps the field of view consistent. The digital camera supply chain that makes a good iPhone module possible is the same one that supplies the consumer market. Beyond those shared basics, the workflow is different and the data product is different.
Try it today
If you want to see what camera-based production monitoring looks like on one of your own lines, the fastest path is to start with one station. A refurbished iPhone, the mount kit, a ring light, and a free account. You will have your first count and downtime data in under an hour, and you will know whether the camera path makes sense for the rest of your plant by the end of the shift.
For the broader picture of what production monitoring covers in 2026, see our production monitoring system guide. For the US shopfloor data collection variant of this same path, see our piece on shop floor data collection.
Start free or join the community to compare camera setups with process engineers running their own lines.
