CLOSE
aigocy@gmail.com

CUP

TWITTER (X) DRIBBBLE LINKEDIN

Configuration

COLORS
CUSTOM CURSOR
Skip to main content

espialtech

Menu
Start a Project

PanelVision

Home
/
Blog
espial
March 18, 2026
No Comments
PanelVision
Visual QC

Automated visual inspection for appliance control panels, comparing every
production unit against a golden template in real time — with zero training data.

Project Details

A global home-appliance manufacturer needed to catch printing, alignment and finish defects on control panels at line speed. The obvious approach — train a deep detector — was the wrong one: panel variants change constantly, and every new SKU would have meant a fresh annotation cycle nobody had time for. We built the system around classical computer vision instead.

A 5MP industrial GigE camera, triggered by a photoelectric sensor synced to the conveyor PLC, captures each panel under a diffused LED ring light and backlight panel that eliminates glare. SIFT and ORB keypoint matching registers the captured image against a reference template; homography alignment corrects for position and angle; and SSIM differencing quantifies the structural deviation. Anything above threshold is localised, flagged and surfaced on a QC dashboard. Inference runs on an NVIDIA Jetson Xavier NX at the line, with no cloud dependency at all.

DELIVERABLES
Computer vision Edge GPU deployment On-site calibration
INDUSTRY
Manufacturing / Consumer Electronics

Project Research

The core research question was whether a template-comparison approach could match a trained detector on a defect class this varied. It could — and it removed the entire annotation bottleneck.

The engineering effort moved from data labelling to the physical layer: our engineers specified the optics, designed the illumination to kill glare and shadow, and tuned the trigger timing against the conveyor so that every image lands in the same frame every time. That physical work is what makes the algorithm look easy.

Project Results

PanelVision detects defects with 99.9% accuracy at under 100ms per panel, running entirely on the edge. Because it needs no training data, a new panel variant can be commissioned in an afternoon rather than a labelling sprint. Inspection is fully automated: the line no longer relies on operator attention holding steady through the last hour of a shift, and the QC dashboard gives supervisors a live view of defect rates by variant and by shift.

CastSight
X-ray Inspection
View Project

CastSight
X-ray Inspection

DESCRIPTION
Deep learning on industrial X-ray imagery to catch internal flaws in aluminium castings — 98% detection at 18 FPS, trained entirely on synthetic defects.
DELIVERABLES
X-ray vision pipeline, Synthetic data generation, Edge inference
INDUSTRY
Automotive Manufacturing