Camthink NeoEyes NE301 is a next-generation smart camera that seamlessly integrates Ultra-Low Power Consumption, Edge AI Inference, and Modular Hardware.
At its core, the NE301 is powered by the STM32N6 MCU, featuring an integrated NPU with 0.6 TOPS of computing power. This allows the device to perform complex AI image recognition and inference locally—zero cloud dependency required.
More importantly, these capabilities are delivered through an out-of-the-box, visually configurable, and OTA-ready firmware system. We’ve made it open-source on Github to ensure it’s accessible for everyday users while remaining highly extensible for developers.
Below is a detailed explanation of the firmware architecture and features of the CamThink NeoEyes NE301.
1. An Integrated Edge AI Camera PlatformThe positioning of the NE301 can be summarized in one sentence: by integrating a built-in Wi-Fi AP and web-based UI, AI inference, model management, image debugging, and network configuration are all embedded directly into the hardware, enabling true plug-and-play operation—power on and configure instantly upon connection.
Key features include:- On-device AI inference based on an STM32N6 MCU with a 0.6 TOPS NPU, capable of running models such as YOLOv8.
- Complete configuration and debugging through a built-in Wi-Fi AP and web-based management interface.
- Modular hardware architecture and open-source design, enabling easy secondary development and expansion for developers.
- Support for a complete low-power workflow, including ultra-low-power sleep (standby current of only 7–8 μA), event-based wake-up, and on-device image inference.
The NE301 requires no app installation and no complex on-site network setup. Deployment and usage can be completed with nothing more than a web browser.
- After powering on the device, press the side button briefly for 2 seconds. The blue indicator LED will light up, signaling that the built-in Wi-Fi AP has been activated.The SSID follows the format:
NE301{last six characters of the MAC address}. - Connect to this AP using a phone or computer, then open a browser and enter
192.168.10.10to access the web-based management interface. - The web interface supports:A fixed login account (system-defined username)A default password: hicamthink, which can be changed in the system settings
The side button is designed for quick interaction and debugging:
- Short press (Capture button) → Instantly captures an image and reports the image and related data according to the application configuratio
- Long press (2 seconds) → Wakes up the Wi-Fi AP and turns on the system indicator LED, making temporary connection and debugging easy
- The Wi-Fi AP automatically enters sleep mode after 10 minutes of inactivity to minimize unnecessary power consumption
Powered by the on-board STM32N6 MCU with a 0.6 TOPS NPU, the NE301 supports real-time on-device inference on live video streams:
- Live preview of real-time video with AI detection overlays directly in the web interface
- Millisecond-level response and low-latency inference, making it suitable for edge AI scenarios such as person, vehicle, and object detection
- AI inference is enabled by default at startup, and can also be manually turned on or off via the web UI
The device firmware includes a complete model management workflow, allowing you to get started immediately and iterate quickly:
- A pre-installed YOLOv8 model is included out of the box, enabling instant hands-on experience with AI detection
The web UI supports:
- Uploading custom-trained model packages
- One-click deployment to the device
- Switching between AI models at any time
We provide a step-by-step guide in the Wikithat walks you through training and deploying YOLOv8 models. In addition, developers can quantize and deploy their own domain-specific or research modelson the NE301, turning it into your own dedicated AI camera tailored to your application.
Inference Parameter Hot Reloading and Image-Based ValidationThe NE301 supports hot reloading of inference parameters, allowing you to fine-tune model behavior in real time:
Use simple sliders in the web interface to adjust:
- Confidence thresholds
- NMS thresholds
- Changes take effect immediately, with updated inference results displayed instantly in the browser
In addition, the device includes built-in image upload validation:
- No need to wait for on-device image capture in real-world conditio
- Upload local images directly and run inference using the currently active model
- Ideal for validating new models across diverse scenarios, significantly reducing debugging and iteration time
This approach enables fast, practical validation of model performance on the actual camera hardware, rather than relying solely on theoretical or offline evaluation.
4. Time-Based Capture, IO/PIR Event Triggers, and Remote Device ControlIn the “Scheduled Capture” section of the web interface, you can:
- Enable or disable scheduled image capture
- Configure capture at specific time points (daily or on selected weekdays; up to 8 time points in the current firmware version)
- Configure interval-based capture (by minutes, hours, or days), ideal for periodic inspections and routine monitoring
The firmware supports event-based image capture triggered by hardware expansion IO:
- Image capture can be triggered by external IO level changes (high/low signals)
- Compatible with external devices such as door sensors, radar modules, and PIR sensors
Combined with the NE301’s built-in PIR interface, alarm interface, and 16-pin IO expansion connector, the device can easily be turned into an event-driven camera, enabling scenarios such as “capture when someone approaches” or “capture on motion or trigger events.”
Remote ControlThe firmware also supports remote device control (requires keep-alive wake-up; enabling this feature will increase power consumption):
- Devices can be remotely controlled via MQTT commands to perform actions such as image capture, sleep, or wake-u
Freely configure the server address, port, and device reporting topics, allowing seamless integration with your existing infrastructure.
Secure by DesignBuilt-in SSL/TLS support with CA certificates and private keys, meeting industrial-grade encryption and security requirements.
No Middleware RequiredThe device acts as both a data source and a gateway. AI inference results generated on the device can be pushed directly to your Home Assistant instance or private cloud, with no intermediate middleware needed.
Coming SoonIn future updates, we will continue to expand protocol support, adding HTTP/HTTPS and WebSocket, enabling integration with even more application scenarios.
6. Hardware management directly in the browserThe NE301 isn’t just about powerful hardware. Its built-in firmware encapsulates complex low-level logic into a visual, out-of-the-box Web UI for seamless configuration.
Image Adjustment: What You See Is What You GetNo coding required. Optimize your image quality in real-time directly through your browser.
- Smart Auto Mode: Enable one-click automatic optimization.
- Precision Tuning: Fine-tune Brightness and Contrast with a 0–100 slider.
- Universal Mounting Support: Built-in Horizontal/Vertical Flip ensures your view is always upright, whether the camera is mounted normally or upside down.
To handle complex outdoor lighting, the NE301 firmware offers three intelligent fill-light strategies:
- Scheduled Mode: Precise control based on time (e.g., 8:00 PM – 6:00 AM), just like a smart streetlight.
- Active-Trigger Mode: Automatically switches on only when an image or video stream is active, balancing image quality with energy efficiency.
- Ultra-Power-Saving Mode: Keeps the lights off—perfect for scenarios requiring extreme standby longevity.
Whether for home coverage or remote wilderness monitoring, the NE301 provides a rock-solid communication backbone:
- Smart WiFi Memory: Automatically remembers and reconnects to the last SSID, featuring intelligent auto-reconnection on signal loss.
- Deep Cat-1 Customization: No more tedious AT commands. If equipped with a cellular module, you can intuitively configure APN, PIN codes, and authentication methods via the Web UI while monitoring real-time signal strength.
Rock-solid foundation, business-focused results. At the firmware level, the NE301 has already built the complete bridge from image capture and environmental adaptation to data transmission.7. Dual-MCU Driven Multi-Stage Power Management
The NE301 is not your traditional “always-on” power hog. Instead, it is a smart camera designed around an “On-Demand” operational logic. By leveraging the synergy between the STM32U073Kx (Ultra-low-power) and the STM32N6 (High-performance AI), it achieves an extraordinary leap in energy efficiency.
Core Architecture: Granular Power ControlWe have integrated a dedicated STM32U0 power management chip to act as the system’s “Sentry.” This chip manages the wake-up and sleep cycles of the primary STM32N6 processor, enabling:
- μA-Level Deep Sleep: Extreme silence to preserve battery life when no action is detected.
- Millisecond Wake-up: Near-instant response to ensure no critical moments are missed.
To allow the NE301 to truly “thrive” in outdoor or unattended environments, we offer highly flexible power options:
- Mobile Deployment: Integrated battery compartment + Solar panel support.
- Fixed Installation: Continuous Type-C power or PoE (Power over Ethernet).
Design Philosophy: It is a high-performance AI camera only when you need it; the rest of the time, it consumes virtually zero power.8. OTA Updates & Configuration Management: Built for Scalable Deployment
In the “Import & Export” section, the NE301 firmware provides a robust toolchain designed for industrial-grade deployment and maintenance.
Modular OTA Updates: Precision and ReliabilityInstead of bulky monolithic updates, the NE301 supports independent updates across four core components:
- APP (Application) | Web (Frontend) | Config (System Configuration) | AI Model
Key Features:
- Partial Updates: Upgrade specific modules without a full system reboot
- Config Retention: Option to keep current settings during the update process.
- Anti-Brick Mechanism:Automatic rollback on failure. If an update is interrupted, the device reverts to the previous stable version, ensuring zero downtime.
Easily upload and replace AI model packages with integrated parameter descriptions. Deploy new models to the edge instantly and verify performance in the debug console with zero friction.
Configuration Backup & Batch Cloning- Standard JSON Format: Export system configurations to
.jsonfor easy backup and version control. - Batch Provisioning: Preview configuration overrides before importing. This makes it effortless to replicate settings across dozens of devices in minutes.
For teams focused on real-world projects and large-scale maintenance, these aren’t just features—they are essential productivity drivers.Conclusion
NE301 — Not Just an AI Toy, But a Professional Edge ToolThe NE301 is meticulously engineered to bridge the gap between “experimental tech” and “real-world deployment.” Its firmware ecosystem caters to every stakeholder in the AIoT lifecycle:
For General Users: Plug-and-Play Simplicity Connect to WiFi, open your browser, and start seeing AI in action. No coding, no CLI—just intuitive controls and instant results.
For Developers & Integrators: Limitless Scalability Benefit from open-source firmware, MQTT/MQTTS connectivity, and modular OTA. With rich hardware interfaces (IO/PIR/Cat-1), it’s built to evolve with your project.
For AI Specialists: End-to-End Control Own your pipeline with ONNX-to-STM32N6 quantization scripts. Hot-swap models via the Web UI and perform real-time A/B testing with instant visual feedback and log tracing.
For Maintenance Teams: Enterprise-Grade Reliability Batch provisioning, fail-safe OTA, and ultra-low power management ensure that once deployed, your fleet stays online with minimal overhead.
The CamThink NeoEyes NE301 is not just another “AI toy” that can barely run a demo. It is a production-ready Edge AI device, designed from Day One for Mass Production, Maintainability, and Scalability.
Comments