CamThink Launches NeoEyes NE302:
Compact STM32N6 Vision AI for Device Integration

September 10, 2026CamThink Team

CamThink today announced NeoEyes NE302, a compact vision AI node designed for developers, OEM product teams, and system integrators adding local visual intelligence to terminals, equipment, workstations, and custom devices.

NE302 brings the essential parts of an embedded vision workflow together on one platform: image capture, hardware encoding, local AI inference, trigger logic, storage, wireless communication, and result delivery. It gives teams a practical way to validate the camera, model, and data path before committing to a custom hardware design.

Built around the STM32N6 microcontroller, NE302 combines an 800 MHz Arm Cortex-M55 processor with the ST Neural-ART accelerator, delivering up to 0.6 TOPS of INT8 AI performance. A 4 MP high-sensitivity image sensor supports full-resolution JPEG capture and H.264 video up to 1080p at 30 fps.

NeoEyes NE302 compact STM32N6 AI vision camera with a white enclosure and external antenna — NeoEyes NE302 packages local vision AI, wireless connectivity, storage, and device-management functions in a compact indoor camera.

From an AI Model to an Embedded Vision Feature

Adding vision AI to a product involves more than running a model on a development board. Product teams must also manage imaging, model packaging, triggers, storage, networking, result formats, firmware updates, and physical integration. NE302 is designed to bring those elements into one testable workflow.

The 38 × 38 mm Main Board carries the camera, STM32N6 vision processor, STM32U0 low-power controller, 32 MB PSRAM, 64 MB Flash, and wireless hardware. The companion Interface Board adds USB Type-C power, MicroSD storage, programming access, and service connections. Together, the boards fit in a 42 × 42 × 20 mm enclosure for compact indoor installation, while the Main Board can also be evaluated for custom device integration.

NE302 Launch Highlights

MCU Edge AI: STM32N6 Cortex-M55 with Neural-ART acceleration up to 0.6 TOPS INT8.
Imaging: 4 MP image capture, hardware JPEG encoding, and H.264 video up to 1080p at 30 fps.
Compact hardware: 38 × 38 mm Main Board and 42 × 42 × 20 mm indoor enclosure.
Memory and storage: 4.2 MB on-chip SRAM / NPU RAM, 32 MB PSRAM, 64 MB Flash, and MicroSD support.
Connectivity: 2.4 GHz Wi-Fi 6, BLE 5.3, MQTT/MQTTS, Webhook, RTSP, and RTMP.
Open development: Open engineering project with application, Web, model, WakeCore, build, flashing, and OTA resources.

A Two-Board Platform for Evaluation and Integration

The Main Board concentrates the image and AI pipeline in a small footprint. The Interface Board gives developers access to power, storage, serial and dual-MCU programming paths without requiring a separate carrier design for first evaluation. This arrangement supports both complete-device testing and deeper firmware or hardware development.

NeoEyes NE302 Main Board component map with STM32N6, STM32U0, memory, Flash, camera, and wireless blocks — The NE302 Main Board carries the vision-AI core, while the Interface Board provides power, storage, controls, and development access.

NeoEyes NE302 Interface Board programming interface map for N6, U0, UART, and boot controls — The NE302 Main Board carries the vision-AI core, while the Interface Board provides power, storage, controls, and development access.

Continuous Vision and Event-Triggered Workflows

NE302 supports continuous USB Type-C power for live preview, video, inference, and device-side processing. A dedicated STM32U0 always-on domain also supports low-power sleep and wake-up control, allowing developers to configure event-driven workflows around PIR or IO input, remote MQTT commands, and scheduled capture.

The device can store visual records on MicroSD or internal Flash and apply upload, retention, and retry policies. This makes it possible to keep evidence locally, send results immediately, or recover from temporary network interruptions according to the application design.

Model Validation and Result Delivery From the Web Console

The built-in Web console provides live preview, image capture, model upload, fixed-image validation, trigger configuration, stored records, and firmware management. Developers can download an NE302-compatible model package from the CamThink Model Zoo, load it through the browser, and check an inference result before connecting the device to an external system.

Structured results can be sent through MQTT/MQTTS or Webhook, while RTSP and RTMP support video workflows. Model compatibility and application performance depend on model architecture, INT8 quantization, supported operators, input dimensions, memory use, firmware, and the tested scene. CamThink recommends validating the target model, lens, lighting, and output path as part of every proof of concept.

NeoEyes NE302 Web console model validation workflow — The NE302 Web console helps teams preview the camera, load a compatible model, and validate inference with a fixed image.

Designed for Terminals, Equipment, and Compact Vision Nodes

NE302 is intended for indoor, device-level vision tasks where the camera needs to capture an image, make a local decision, and deliver a result to an existing product or application. Potential project directions include:

Self-service terminals: user-presence detection, barcode or QR workflows, and terminal-state recognition.
Access and entry devices: event capture, person detection, passage-state analysis, and interaction logic.
Equipment and workstations: fixed-view state detection, object counting, and visual process checks.
Displays and panels: local reading of counters, status indicators, or equipment interfaces with a validated model and imaging setup.
Embedded AI prototypes: evaluation of STM32N6 models, lenses, triggers, storage policies, and result-delivery methods.

These are reference directions rather than preconfigured industry applications. Production use requires project-specific validation of the model, field of view, target distance, lighting, network behavior, and acceptance criteria. For outdoor, weather-exposed, cellular, or PoE installations, CamThink offers other NeoEyes deployment paths.

Open Engineering Resources

NE302 launches with a public GitHub repository, technical documentation in the CamThink Wiki, downloadable models in the CamThink Model Zoo, and versioned firmware packages through GitHub Releases.

The source project separates the first-stage bootloader, application, Web interface, model package, wireless firmware, and WakeCore components. Developers can build, package, flash, and update individual components using the documented toolchain. Model packages and application firmware must use compatible ST Edge AI runtime versions.

Availability

NeoEyes NE302 is available to order through the official store, with a standard price from USD 109.0. Configurations include 2.4 GHz Wi-Fi 6 with BLE 5.3, Wi-Fi HaLow for longer-range sub-GHz connectivity, and 88° or 137° horizontal field-of-view lens options.

Developers and product teams can contact CamThink to discuss model compatibility, lens selection, wireless configuration, firmware development, evaluation samples, or custom Main Board integration.