What does AI Object Matching do?

It runs a trained neural network to detect objects in camera images and reports each one with its class, position, orientation and score – also for mixed, varied or hard-to-describe parts where rule-based matching struggles.

Where does the model come from?

You train it on your own parts with LumiScan Training, using images annotated with LumiScan Annotation. The result is a standard ONNX file. No CAD model is needed.

Which hardware does it run on?

On ctrlX OS – optionally accelerated on a Hailo TPU in a ctrlX CORE X3 or X7 – or on a Windows 11 or Ubuntu 24.04 PC with an NVIDIA GPU.

How many licences do we need?

One licence per network executed in parallel.

Where does the app get its images from?

It does not talk to a camera itself. It reads the image node of another app on the same device, normally LumiScan Cam Driver for GenICam cameras. Supported pixel formats are Mono8, Mono12, RGB8, RGB12, BGR8, BGR12 and the common Bayer 8/12-bit formats.

LumiScan Ecosystem: AI Runtime

LumiScan AI Object Matching

ONNX-based object detection, matching, and pose estimation at runtime.

The LumiScan Object Matching Runtime is a software component for executing ONNX-based AI networks for industrial object detection, object matching, and pose estimation.

The runtime provides a standardized execution environment for trained object matching models and enables reliable detection as well as determination of position and orientation of objects in 2D and 3D image data.

The software acts as a container for ONNX-compatible neural networks and can be used in conjunction with LumiScan Vision and Robotics solutions. Typical applications include the identification of components, determination of gripping positions, and object localization in automated manufacturing and logistics processes.

The runtime is intended for productive deployment on machine and target systems.

Scope of Performance

Execution of ONNX-compatible object matching models
Object detection and object classification
Determination of object position and orientation
Support for oriented object matching
Processing of 2D and 3D image data
Integration into industrial automation and robotics systems

Functions

ONNX Runtime Container
Inference of trained neural networks
Pose Estimation
Rotation and pose determination
Multi-object detection
Real-time capable execution for industrial applications

Typical Applications

Vision Guided Robotics
Bin Picking
Machine Tending
Automated Assembly
Pick-and-Place applications
Part identification
Object localization in chaotic scenarios
Quality control and presence inspection

Compatibility

Compatible with:

LumiScan VGR Runtime
LumiScan VGR Lite Runtime
LumiScan Engineering
LumiScan Metrology SDK
LumiScan Vision Suite
ONNX-compatible neural networks

License Model

Runtime License
Per system / machine
Perpetual License
Optionally with Software Maintenance

Scope of Delivery

LumiScan Object Matching Runtime License
ONNX Runtime Container
Interfaces for integration into LumiScan applications
Documentation

Not Included

Training of neural networks
Creation or optimization of training data
Development tools for model creation
Individual AI models
Individual integration services
Software support (unless separately contracted)

All apps in the ecosystem