Visual quality inspection
Check surfaces, completeness, assembly conditions and markings against agreed classes.
Data & AI Engineering
Computer Vision: Convert image data into verifiable decisions
From visual inspection to recognition and anomaly detection, we engineer the complete system: image capture, evaluation, model deployment and integration into operations.
The challenge is not just the tool
Manual inspection and automated vision systems both depend on lighting, camera placement, material variation and the definition of an acceptable result. A demo model is not proof of reliability in a production line.
We define the inspection decision first: which defects matter, what false alarms cost and how the outcome changes the workflow. That guides the imaging setup, dataset and technical integration.
What we engineer
Detection and classification
Recognize objects, components and relevant operational states.
Segmentation and measurement
Identify image regions and assess geometry with appropriate calibration.
Anomaly detection
Detect deviations when exhaustive defect examples are unavailable.
Production system integration
Transfer decisions to MES, ERP, controls or specialist applications with traceability.
Architecture and trade-offs
Imaging hardware, optics and lighting are part of the system. A better neural network cannot compensate for every capture problem. We evaluate image conditions, annotation rules and representative datasets together.
We measure false positives, false negatives, throughput and robustness across real variations. Inference may run centrally or at the edge depending on latency, network availability and security constraints. High-impact decisions need explicit review paths.
From the requirement to production
01 · Define the inspection rule
Agree on defects, thresholds, throughput and response.
02 · Capture representative images
Review cameras, lighting, materials and process variation.
03 · Establish a baseline
Compare classical image processing and trained models.
04 · Integrate the complete pipeline
Build capture, inference, decisions and system interfaces.
05 · Monitor in production
Track capture quality, error rates, drift and process changes.
Related capabilities and use cases
Run vision inference on local devices and production lines.
Apply vision to inspection and quality records in automotive.
See practical implementation requirements for mid-sized teams.
Frequently asked questions
How many training images are required?
The number depends on variation and the defect classes. Representative coverage and an independent test set are more important than an arbitrary minimum.
Can visual inspection become fully automated?
Not in every case. The appropriate level of automation depends on failure risk and quality requirements.
Must the system run in the cloud?
No. Edge inference can be preferable for latency, connectivity or data restrictions.
How do we assess success?
Using meaningful error metrics, realistic data and measurements in the actual operating process.
Technical discovery
Which visual inspection decision needs to be more reliable?
We review current systems, technical boundaries and a practical first implementation step.
Aleksandar Basara
Managing Director
aleks@valanor.co
+49 (0) 711 16 44 6 83
Lindenspürstrasse 22, 70176 Stuttgart
