OCR vision sensor PV4

smartcolorfor detection of wafer cracks and defects

Description
The Vision Sensor PV4 is an ultra-compact, freely programmable AI vision sensor combining a hexa-core Arm® Cortex-A55 CPU with an integrated neural processing unit (NPU) delivering 2 TOPS of on-device AI inference. It executes both classical image processing algorithms and AI models entirely on the device without additional hardware or proprietary runtimes. The PV4 features a modular sensor design and integrated lighting options, enabling flexible adaptation to different optical requirements and dynamic production environments.

Mechanical variants

• Cubic: 50 × 50 × 40.8 mm, 205 g
• Matchbox: 45 × 53.7 × 25.5 mm, 125 g

Advantages

• Ultra-compact & easy to integrate: two small housing variants for confined installation spaces and retrofit projects.
• Freely programmable like a Linux industrial PC: open Linux system; users can deploy scripts (e.g. Python, C++) and integrate AI models; avoids rigid function packages.
• On-device AI with integrated NPU: 2 TOPS NPU enables local AI inference and combination with classical image processing for high-performance evaluation at the data source.
• Modular sensor design: selectable sensor heads (e.g. 2064 × 1552 px or 2472 × 2064 px, color or mono) that can be adapted to new tasks.
• Integrated lighting & autofocus option: Matchbox variant features four synchronized high-intensity LEDs; optional liquid lens autofocus compensates for rapid distance changes.
• Fast integration: 2.5 Gbit/s Ethernet, opto-isolated I/Os (2× In / 4× Out), trigger interfaces and optional modules such as UIM and I/O Expansion Box.
• Efficient parameterization & fast onboarding: ViewIT enables intuitive setup of classic image processing and AI combinations; optional HALCON support for advanced users.

Applications
The PV4 is suited for industrial tasks that require modular sensors, AI capability, and compact form factor. Typical application areas include:

• Pharma: reliable code and label reading with changing batches and small fonts; AI-based segmentation increases recognition reliability with varying layouts.
• Food & Beverage: inspection of fill levels and labels on changing packaging shapes and reflective surfaces; AI supports varying colors and textures.
• Production and Web Inspection: detection of surface defects such as cracks or structural deviations by combining AI and classical processing.
• Packaging & Logistics: barcode and label validation on damaged or partially covered packages; AI-based classification supports sorting and routing.
• Printing: inspection of print quality, color deviations and positioning; combines deterministic inspections with AI for varying layouts.

Target group
Machine builders, OEMs and system integrators who need a powerful, compact and freely programmable AI vision sensor. Suitable for users who require more flexibility than classic vision sensors but prefer lower complexity than an AI industrial PC. The PV4 supports running custom algorithms and AI models directly on the camera as well as preconfigured tools for less expert users.

Documentation (available titles on the product page)

• Product Specification Matchbox
• Product Specification Cubic
• Hardware Manual
• Getting Started
• SDK Documentation
• 3D File

FAQs (selection)

• What differentiates the Vision Sensor PV4 from classic vision sensors? The PV4 is a fully programmable, AI-enabled platform offering greater flexibility and performance than classic vision sensors; suited for frequent format changes or additional inspection steps.
• Which AI models can the Vision Sensor PV4 run? The integrated NPU (2 TOPS) supports on-device AI inference. Models for classification, segmentation, OCR and similar tasks can be integrated and adapted via the Linux system.
• Which variants are available? Two housing types (Cubic & Matchbox) and several sensor variants are available; selection depends on installation space, lighting and required resolution.
• Does the PV4 require additional hardware for AI? No. Classical and AI-based image processing tasks run entirely on the device supported by the ARM CPU and integrated NPU.
• How is software developed on the PV4? Users can program freely (e.g. C++ or Python) or use ViewIT and optional HALCON; the open Linux architecture allows integration of custom scripts, models and libraries.

Technical specifications

• CPU: hexa-core Arm® Cortex-A55
• NPU: integrated neural processing unit, 2 TOPS
• On-device processing: supports both classical image processing and AI models locally
• Mechanical variants: Cubic (50 × 50 × 40.8 mm, 205 g); Matchbox (45 × 53.7 × 25.5 mm, 125 g)
• Sensor variants: examples 2064 × 1552 px or 2472 × 2064 px; available in color or mono
• Integrated lighting: Matchbox variant with four synchronized high-intensity LEDs
• Optional autofocus: liquid lens option for rapid distance compensation
• Connectivity: 2.5 Gbit/s Ethernet; opto-isolated I/Os (2× In / 4× Out); trigger interfaces
• Optional modules: UIM (Universal Illumination Module), I/O Expansion Box
• Software: open Linux system; programmable in Python, C++; ViewIT for setup; optional HALCON