OCR built for the
factory floor.

Read serial numbers, batch codes, expiry dates, and labels at full line speed — even on curved bottles, embossed metal, dot-peen markings, or low-contrast surfaces. Industrial OCR that doesn't break when consumer OCR does.

OCR reading batch number, manufacturing date, and expiry date from a pharmaceutical bottle on a conveyor line

Traceability without slowdowns.

Industrial OCR is harder than the document OCR most engines were trained for. Codes are stamped, etched, dot-peened, or printed onto curved, reflective, or moving surfaces under variable lighting. Consumer OCR fails. MachVizyon's OCR uses CNN-based detection and transformer-based recognition tuned on industrial datasets — handling embossed and engraved characters, ink bleed, partial occlusion, and skew without manual tuning.

Read every code, link it to your MES batch record, verify it matches expected format and value, and trigger automatic rejection or routing — all in under 30 milliseconds per part.

Reads what other systems can't.

Engineered for the surfaces, speeds, and lighting of real production lines.

🔤

Multi-format support

Inkjet print, laser etch, dot-peen, embossed, engraved, hot-stamped — single model handles all marking methods.

🌀

Curved & rotated text

Reads codes wrapped around bottles, cans, and cylindrical parts — auto-rectifies skew and curvature.

✓

Format validation

Define expected format per station (regex, checksum, date range). Mismatches flagged before product ships.

🔍

Low-contrast read

Handles white-on-white, embossed, or near-invisible markings via specialized lighting and contrast enhancement.

🔗

MES / ERP linkage

Each read is timestamped, linked to product, and posted to MES — enabling end-to-end batch traceability.

📑

Codes & barcodes

Reads alphanumeric, Data Matrix, QR, GS1-128, and Pharmacode in the same pass — no separate scanner required.

Trace every unit, every batch.

Wherever regulatory traceability or quality genealogy matters.

💊

Pharmaceutical serialization

Verifies GTIN, batch, expiry, and serial codes on every vial, bottle, and carton. DSCSA / EU FMD compliant.

🚗

Automotive part marking

Reads engraved or dot-peened part numbers and Julian dates on engine, transmission, and brake components.

🍺

F&B date coding

Verifies inkjet best-before dates and lot codes on bottles, cans, and pouches at 800+ units/min.

🔌

Electronics traceability

Reads laser-marked codes on PCBs, semiconductor packages, and connector housings down to 1mm character height.

🏗️

Steel coil & billet ID

Reads paint-sprayed or stenciled heat numbers and coil IDs on steel from rolling mill to dispatch.

📦

Logistics & carton verify

Reads shipping labels, AWBs, and carton codes at sortation hubs — complementing existing barcode scanners.

Detect text. Rectify. Recognize. Validate.

Two-stage pipeline that handles industrial challenges other OCR systems collapse on.

1

Trigger & capture

Encoder or photo-eye triggers camera. Frame captured at controlled exposure under specialized lighting.

2

Text detection

CNN locates text regions with rotated bounding boxes — handles tilted, curved, and stacked text.

3

Rectify

Each region is unwarped and normalized — curvature flattened, skew corrected, lighting equalized.

4

Recognize

Transformer-based recognizer outputs character sequence with per-character confidence scores.

5

Validate & act

Read checked against format/value rules. Mismatches trigger reject; valid reads posted to MES.

Industrial accuracy at line speed.

Numbers from active pharma, automotive, and F&B deployments.

99.6%
Read accuracy
Character-level accuracy across embossed, etched, and inkjet codes.
< 30ms
Per-part latency
End-to-end from trigger to validated read on edge GPU.
800+
Units per minute
Sustained throughput on pharma vial lines with 100% read coverage.
0.4%
No-read rate
Fraction of parts requiring manual review — down from 6%+ on legacy OCR.

Drop-in replacement for legacy OCR.

Hardware and integration paths chosen to work alongside the systems you already run.

📷 Imaging hardware

  • Area-scan and line-scan industrial cameras (Basler, Hikrobot, Allied Vision)
  • Telecentric lenses for embossed/engraved character reading
  • Dome, dark-field, or coaxial lighting per surface type
  • Trigger via encoder, photo-eye, or PLC signal
  • Multi-camera arrays for 360° read on cylindrical products

🖥️ Compute

  • NVIDIA Jetson Orin / Orin Nano edge appliances
  • TensorRT-optimized models for FP16 inference
  • Hot-standby pair option for high-availability lines
  • Local result buffer if MES connection drops
  • Operates -10°C to +55°C, IP65-rated enclosures

🔌 Integration

  • OPC-UA, Modbus TCP, Profinet for PLC reject signaling
  • REST & gRPC APIs for MES and ERP write-back
  • Native plug-ins for SAP ME, Siemens Opcenter, Rockwell FT
  • Image archive with read overlay for audit (S3 / on-prem)
  • Track & trace export to GS1 EPCIS / Tracelink

📅 Timeline

  • Week 1 — sample collection, lighting & lens trial
  • Week 2 — model fine-tune on your code formats
  • Week 3 — pilot install & PLC/MES integration
  • Week 4 — IQ/OQ/PQ validation (regulated industries)
  • Ongoing — accuracy reports & periodic model refresh

Send us your toughest codes.

Email a few photos of the markings you want to read — embossed, etched, low-contrast, curved, whatever — and we'll send back live OCR results within 48 hours.

Request a demo