We build custom perception pipelines for industrial customers — 3D point-cloud classification, object detection, anti-collision and worker-safety monitoring. Five years of LiDAR engineering in active production sites, packaged as software you can deploy on your own plant.
Industrial perception AI is software that turns raw 3D point clouds and 2D camera images into machine-readable decisions on a plant floor. We design, build and ship those pipelines for industrial customers — scrap-vs-slag classification, crane and vehicle detection, worker safety in ATEX zones, material-flow optimisation.
We started with one job: measuring stockpiles with LiDAR inside an active mining operation. Five years and forty live installations later, we have a deep toolbox for industrial point-cloud and computer-vision work — Riegl, Livox and Faro on the hardware side; PCL, Open3D, OpenCV and PyTorch on the software side; OPC UA, REST and MQTT into your control system. We bring that toolbox to your data.
We don't sell a generic AI platform. Every engagement starts with a short data audit: what sensors do you have, what does your data look like, what decision are you trying to automate? From there we propose a fixed-scope pipeline — typically 4–12 weeks from contract to live tool. For larger programmes we move into a monthly retainer model with shared road-mapping. Either way you own the code: see our FAQs on IP and hardware reuse for the standard terms.
Production references across salt, sugar, copper, charcoal and minerals.
Scrap-vs-slag classification on steel-mill yards, ore-grade detection on conveyors, mixed-material search and segmentation on stockpiles. Point-cloud classifiers and 2D vision models trained on your material — not a public benchmark.
Crane position + anti-collision in waste bunkers and warehouses, vehicle detection at plant gates and weighbridges, railcar and wagon recognition for loading automation. LiDAR-based, lighting-independent, works in dust.
Worker detection inside ATEX zones and around moving plant equipment, safety-distance monitoring on automated cranes and AGVs, hazard-zone clearance checks before a process restart. Auditable event logs included.
No long sales cycles. We scope every engagement around your data — start with a workshop, move to a fixed-scope project, optionally into a long-term retainer.
A 1–3 day on-site or remote workshop. We audit your sensors, your data pipeline and the decision you want to automate, and ship a written feasibility note with a recommended next step. Starting at a few days. Fixed price.
A 4–12 week project that delivers a working perception pipeline — data parser, detection/classification algorithm, dashboard or OPC UA integration into your control room. Fixed scope, fixed price, you own the code at handover.
A monthly retainer for ongoing model retraining, sensor migrations, new use cases and on-call support. The right fit when you want a perception engineering team without hiring one. Quarterly road-mapping.
All three tiers include source code, documentation and a written handover — no black boxes, no licence dependence on us.
Distinguish ferrous scrap from cooled slag on a steel-mill yard, in real time, from a portal-mounted LiDAR + camera pair. Feeds the EAF charging schedule and reduces sorting labour on the yard. Read the deep dive →
Continuous classification of bulk material on conveyors and at transfer points — feeds blending models, prevents off-spec batches and gives the control room a real material-balance instead of an inferred one. Read the deep dive →
LiDAR tracks crane bridge, trolley and hook position in a waste bunker or warehouse. Drives anti-collision interlocks, optimises grab routing, integrates with the crane PLC over OPC UA. See OWL EYE® Waste Bunker for a productised version. Read the deep dive →
Truck, loader and forklift recognition at plant gates, in-plant intersections and weighbridges. Drives gate control, traffic-light interlocks and in-plant routing — works at night, in fog and in dust. Read the deep dive →
Detect, count and identify railcars at a loading station. Drives automated loading sequences, prevents misroutes and feeds a per-wagon material log into ERP. Read the deep dive →
Person detection in hazard zones — ATEX areas, swing radius of cranes, conveyor pull-cords. Triggers slowdowns, hard stops or audible warnings. Auditable event log for the safety officer. Read the deep dive →
We keep the stack boring on purpose. Every library on this list has a multi-year track record in industrial production environments.
Single raw 3D point-cloud scan in, six lid centerpoints plus the wagon number out. Built as a stateless six-stage geometric pipeline (NumPy + scipy) so the core ports cleanly to C++/C# for the final controller. Wrapped in an Electron + Vue 3 + Three.js desktop UI for the plant operators. Scan processed in under two seconds, mean centroid error under five centimetres against hand-labelled ground truth.
The Electron UI as the plant operator sees it: load a freshly arrived scan, the algorithm classifies the point cloud into ground, wagon body and lid cluster, projects centerpoints and updates the train log. All stateless on the algorithm side — the tracker is a separate state machine.
How we scope a project →Sensor setup, sample data, the decision you want to automate — anything you have. We come back within two business days with an honest first assessment and a fixed-price scope for the Discovery workshop.