Computer vision

Your production task. A visible result.

Start with what you need to inspect or measure. Explore the approach, a process diagram and project evidence.

Solutions and projects

Choose a task to see the available approaches

Project stages differ: site deployments, pilots, applications and research. We agree readiness and acceptance criteria for your site separately.

Hygiene entry control

Face and chest capture at the kiosk, followed by washing and hand treatment checks. Access opens after the required checks. An anonymised two-zone pilot layout.

Packaging damage inspection

An overhead camera captures the pouch under diffuse and low side lighting. The images help inspect film tears, seal damage and deformation before shipment.

Product identification in crates

Identify products and empty crates on a conveyor. The referenced deployment reports about 0.2 seconds to deliver the result; throughput is validated for each setup.

Beef marbling assessment

An optical device with a camera and controlled lighting for beef assessment. RibeyeAnalyzer displays the cut image, segmentation and quality parameters for the operator.

Operator holding a phone with the real MarbScan analysis in a production area

Real MarbScan screen with the meat boundary, fat highlights and assessment

MarbScan — mobile assessment

Android app for pork marbling and colour assessment at intake, using a separate six-class scale. Analysis runs offline on the phone. Adaptation to your data is agreed for deployment.

Video monitoring of vaccination

Pilot project. A revised three-camera layout observes the neck, two preparation tables and operators. Action sequences support review of vaccination, needle handling and refilling, with local staff tracking.

Pig counting and video weight estimation

Contactless livestock counting and weight estimation from passageway video. Each animal is detected, tracked and assessed to monitor daily weight gain without moving it through scales.

Livestock counting from aerial imagery

MooCowsCount desktop application for counting cattle, sheep, goats and pigs in drone imagery. AI assisted annotation with Excel export. Adaptation to your data is agreed for deployment.

Big bag and container counting

Video counting of sacks, big bags, containers and crates in warehouses and production areas. Real-time tracking with an inventory dashboard. The frame shows bags on a loading conveyor.

MooMind Scan — pallet barcodes

An Android application for reading SSCC, EAN-13 and DataMatrix codes on mixed pallets. Our lightweight barcode segmentation model and industrial OCR support Cyrillic text. Adaptation to your data is agreed for deployment. The image illustrates the approach.

Production analytics

Dashboards and integration of computer vision systems with MES, ERP and 1C. Real-time line metrics, report export and BI visualisation of key KPIs. The image illustrates the approach.

Leather defect inspection and grading

R&D for leather grading: one camera captures four images as the side lights switch on sequentially. Colour and surface relief provide complementary information for defect inspection and grades 1/2/3/4/F.

Ultrasound milk yield prediction

Research and development. A non-invasive milk yield forecast for heifers in 5 minutes using an udder ultrasound scan. Supports early planning of the dairy herd.

Have another task? Describe the process so we can assess a suitable approach.

How we work Industries

Project evidence

Measurements, methodology studies and pilot targets are identified separately. Customer names are withheld; evaluation methods can be discussed on request.

Suspended optical device at the carcass inspection point

Real RibeyeAnalyzer interface with the captured image and segmentation result

Marbling gradingVertically integrated agricultural group

An objective marbling scale for consistent grading

5 grades30 samples3,600+ images

30samples in the methodology study

5 gradesobjective grading scale

Research institutevalidation against laboratory fat

Task, approach and evidence

ChallengeMarbling was assessed by eye, subjectively and without a consistent scale. This led to grade disputes, lost margins on premium raw materials and difficulty standardising sorting.

ApproachThe optical device captures the exposed beef cut under controlled lighting. Image analysis estimates fat inclusions. A five-class methodology was compared with laboratory fat measurements on 30 samples and more than 3,600 images.

ResultA five-class beef assessment methodology and an operator interface. This study describes beef grading; it does not establish the accuracy of the separate six-class MarbScan pork application.

In the catalogue

Computer vision for hygiene entry control

The employee stops at the kiosk for a face and chest image. Cameras check the washing sequence and hand treatment before access opens. A pilot is deployed; at least 95% is the contractual accuracy target.

Product and empty-crate identification on the line

Process illustration: camera and lighting above a conveyor. It illustrates the inspection setup, not measured model output.

Conveyor identificationMeat processing plants

Product and empty-crate identification on the line

80+ SKUsOPC / ERP / scales

99,4 %SKU recognition accuracy

OPC · ERPintegration + weighing terminal

80+SKUs in production

Task, approach and evidence

ChallengeManual product identification on conveyors involved dozens of SKUs, a risk of misclassification and difficulty detecting empty crates at line speed.

ApproachOur computer vision detectors recognise products in crates and detect empty crates with a GPU optimised runtime. The system integrates with the plant's OPC server, ERP and weighing terminal. The server can be hosted by the plant or MooMind. SKU recognition accuracy exceeds 99.4%; the interval from crate stop to MES data delivery is 200 ms.

ResultAutomatic product identification and empty crate checks on the line, integrated with business systems. Previously, an operator selected the SKU on a touchscreen, taking 30 seconds to several minutes and introducing a risk of misclassification and waste. The system has been deployed at several plants with individual configurations.

In the catalogue

The crate project reports over 99.4% identification accuracy and about 200 ms from a stopped crate to MES delivery. These figures belong to that deployment, not a guarantee for another line; confirm the evaluation protocol for your setup.

Let's discuss your project

Describe what you need to check or measure. We will clarify the process, available data and a suitable next step.