Author

Andreas Ohrvall

CTO · LexData · 41 articles

A well pad from above, separator, scrubber and pig launcher boxed, from a customer drone survey

Industries · 5 min read

Checking a well pad's own equipment list from a drone pass

A box on the separator, the scrubber and the pig launcher read against the pad's build plan, and the skid the correction rate catches first.

Andreas Ohrvall · Nov 10, 2026

Edge box beside a video recorder in a plant control cabinet, cable and recorder boxed, generated scene with detections from our model

Computer vision · 6 min read

How to choose an object detection model architecture for a camera on your own site

Start from the decision the plant has to make, then the box beside the recorder the model has to fit, and only then the family. The benchmark comes last.

Andreas Ohrvall · Oct 4, 2026

Loading dock from a mounted camera, trucks at the bays and a forklift with a pallet boxed, generated scene with detections from our model

Labeling · 6 min read

Passive image collection at the edge, letting the camera build its own training set

Leave the dock camera sampling over a weekend, drop near-duplicates at the box, pull the rare scenes with a typed description, and send the rest to review.

Andreas Ohrvall · Oct 3, 2026

Bottling line under the fill head, bottles queued and one missing its cap, generated scene with detections from our model

Computer vision · 7 min read

Traditional computer vision vs deep learning, when a threshold is enough and when the line needs a model

A colour threshold checks caps until the lamp changes. A scratch on dark paint has no rule anyone can write. Most lines end up running both, in that order.

Andreas Ohrvall · Oct 3, 2026

Warehouse aisle with racked pallets from a high camera, forklift and pallet boxed, generated scene with detections from our model

Labeling · 6 min read

Zero-shot detection vs custom training, where a prompted model runs out

A prompted model boxes the forklift first try and has never seen a wafer scratch. Test on a small labeled set, then let corrections train the model that has.

Andreas Ohrvall · Oct 3, 2026

Bottling line under the fill head, one cap missing, generated scene with detections from our model

Operations · 7 min read

Streaming computer vision predictions into the event bus a plant already runs

The capper camera and the labeler camera on a filling line publish each detection as a timestamped event by webhook, and the plant's own bus carries it.

Andreas Ohrvall · Oct 2, 2026

Edge runner beside a recorder, generated scene with detections from our model

Edge · 6 min read

Computer vision on multiple video streams from one runner

Twenty cameras, one box beside the recorder. Fair sampling, a slow camera that drops its own frames, an unplugged one that stalls nobody, one heartbeat each.

Andreas Ohrvall · Oct 1, 2026

Bottling line with bottles and caps boxed under the inspection camera, generated scene with detections from our model

Edge · 7 min read

CPU vs GPU for computer vision inference, and when a CPU is enough

A cap station checked every couple of seconds and a shelf camera sampled every few minutes both run on a CPU. The GPU earns its keep when the load piles up.

Andreas Ohrvall · Oct 1, 2026

Orchard lane with the drivable path and tree rows boxed, from a customer tractor camera

Edge · 6 min read

Deploying computer vision models to edge devices beside the camera

A robot cell that cannot wait for a round trip and an orchard with no uplink. Export by device, run beside the recorder, and get the next version out there.

Andreas Ohrvall · Oct 1, 2026

A compute box beside a video recorder in a plant cabinet, where frames become events, generated scene with detections from our model

Edge · 6 min read

Edge computer vision for industrial automation, where the camera is the heaviest sensor on the plant network

A plant network built for PLC tags was never built for video. The runner beside the recorder turns frames into events, and only events cross the network.

Andreas Ohrvall · Oct 1, 2026

Stamping line with sheets leaving the press, generated scene with detections from our model

Operations · 6 min read

How to increase inference speed for computer vision without a bigger GPU

A defect model that cannot keep pace with the press gets fixed by resolution, sampling and change gating long before the hardware budget is touched.

Andreas Ohrvall · Oct 1, 2026

Thermal camera on a fence line at night, two people at the fence, from a customer site camera

Industries · 6 min read

Perimeter security with fixed cameras, object detection and a drone sent to look

A frame every two seconds is enough to catch a person at the fence, a CPU is enough to run it, and the drone is the second look rather than the detector.

Andreas Ohrvall · Sep 30, 2026

Robot arm loading a CNC machine behind a safety fence, generated scene with detections from our model

Industries · 6 min read

Custom vision models for industrial robots, trained on your own castings

The plant's own castings labeled from the cell's camera, defect, orientation and grasp point as the three questions, and a new variant handled by labeling it.

Andreas Ohrvall · Sep 30, 2026

A robot cell on a plant floor, the kind of line a runner in the cabinet watches, generated scene with detections from our model

Edge · 6 min read

Edge AI fleet management for fifty runners across five plants

Fifty small boxes beside fifty recorders, one model version, and the IT patch that changes what every camera sends. What a fleet needs before the second plant.

Andreas Ohrvall · Sep 30, 2026

A compute box beside a video recorder in a plant cabinet, generated scene with detections from our model

Edge · 6 min read

What computer vision model deployment means after the first week

Deploying a vision model is a decision about where it runs, how a new version replaces the old, and what tells you it has started to be wrong.

Andreas Ohrvall · Sep 30, 2026

Warehouse aisle from a high camera, forklift and racked pallets boxed, generated scene with detections from our model

Industries · 6 min read

AI in robotics after the robot ships, what the warehouse cameras keep learning

The forward camera boxed pallets and people well at the pilot site. Then the racking moved, and the edge cases the planner never saw came back for review.

Andreas Ohrvall · Sep 29, 2026

Small compute box beside a video recorder in a plant cabinet, recorder and cable boxed, generated scene with detections from our model

Computer vision · 6 min read

What the big vision models are for, and what runs on the camera

The big models find the frames, draw the first boxes and tighten them, and a person checks. The small model trained on those labels runs beside the recorder.

Andreas Ohrvall · Sep 29, 2026

Packaging line, cartons passing a scanner on a conveyor, generated scene with detections from our model

Industries · 6 min read

Manufacturing ERP systems and the gap between the ledger and the floor

The WIP module knows what somebody typed at shift end. A camera counting units off the fixture as they happen posts the record by webhook, minutes behind.

Andreas Ohrvall · Sep 29, 2026

Loading dock from a mounted camera, trucks at bays and a forklift with a pallet boxed, generated scene with detections from our model

Industries · 7 min read

Sensor fusion with computer vision, the tag says which pallet and the camera says what happened to it

The RFID portal reads a tag at the dock door. The camera above it sees the pallet lifted, filled or left behind, and catches the tag that fell off on the way.

Andreas Ohrvall · Sep 29, 2026

Bottling line with bottles queued under the fill head, one missing its cap, bottle and cap boxed, generated scene with detections from our model

Computer vision · 6 min read

What YOLO does in one pass and what that costs

One look at the frame is why a single-pass detector fits the box beside the recorder, and why it misses the small cluttered things a two-stage model finds.

Andreas Ohrvall · Sep 29, 2026

Packaging line, cartons with labels moving past the scanner, generated scene with detections from our model

Operations · 6 min read

Computer vision on the industrial HMI, the camera's verdict on the operator screen without the noise

The verdict as a state, the frame one click away, an alarm the operator acknowledges like any other, and everything else kept off the screen.

Andreas Ohrvall · Sep 28, 2026

Edge runner beside the recorder in a control cabinet, generated scene with detections from our model

Labeling · 6 min read

YOLO knowledge distillation, let the big model label and the small model run

A large model boxes the warehouse frames once, a floor lead corrects them, and a compact detector trained on that set runs beside the recorder.

Andreas Ohrvall · Sep 28, 2026

Loading dock from the mounted camera, trucks, forklift, pallet and person boxed, generated scene with detections from our model

Edge · 8 min read

AI cameras vs IP cameras, and what changes when the model moves to the edge

The dock camera has streamed to a recorder for six years. A model can watch that stream in the cloud, on your servers or beside the recorder. No new camera.