No products in the cart.
NeoEyes NE503 4K PoE Edge AI Camera
Price range: $999.00 through $1,199.00
NeoEyes NE503 is a 4K PoE edge AI camera for on-device video analytics. Hailo-15H provides 20 TOPS INT8 compute, with Sony IMX678 imaging, RTSP output and containerized applications. Choose 4 GB or 8 GB LPDDR4 with the 8-32 mm AF zoom lens. IP67 housing supports outdoor installation.
4K video and AI applications on one camera
NeoEyes NE503 combines image capture, local inference and application logic in an IP67 PoE camera. It is built for integrators and developers who need video plus structured AI results at the installation site.
Use RTSP for video recording and monitoring. Run containerized applications to process frames, apply project rules and send events to your systems.
Choose your configuration
Choose the lens for your scene coverage and the memory capacity for your application workload.
Lens options
| Item | AF zoom lens | Motorized zoom lens |
|---|---|---|
| Focal length | 8-32 mm | 2.8-12 mm |
| Horizontal field of view | 44.5° (wide) / 14.5° (tele) | 109.8° (wide) / 46.2° (tele) |
| Optical zoom | 4× | Approximately 4× |
| Lens control | Autofocus and motorized zoom | Motorized zoom and focus |
| Choose for | A narrower view and detail at greater distances | Wider scene coverage, such as entrances and open work areas |
Scroll to view all columns.

Memory options
| Item | 4 GB LPDDR4 | 8 GB LPDDR4 |
|---|---|---|
| Application fit | Workloads with a defined memory budget that fits within 4 GB, including system services | Larger applications or concurrent services that need more memory |
| AI compute | 20 TOPS INT8 | 20 TOPS INT8 |
| Internal storage | 64 GB eMMC | 64 GB eMMC |
Scroll to view all columns.
Run and extend AI applications
Hailo-15H supports local multi-model inference. Pre-installed models include YOLOv8n person detection, Face Landmarks and CLIP Image Encoder. Person detection, occupancy monitoring and safety helmet detection are available as verified application examples. Each requires an installed app and a loaded model.
OCR, recognition and custom inspection workflows require suitable models and application logic. Custom models use HEF format, compiled for Hailo-15H through the Hailo model compilation workflow. Inference speed and concurrent workload capacity depend on the models, input sizes and application load.
4K imaging and RTSP integration
The Sony IMX678 1/1.8-inch sensor and Hailo Gen2 AI-ISP provide 3840 × 2160 imaging, HDR and AI denoising for low-light scenes. H.264/H.265 hardware encoding supports up to 4K at 30 fps. Main, sub and third RTSP streams connect to compatible VMS and NVR systems.
Video encoding resolution and frame rate are separate from AI model input resolution and inference speed.
Deploy and integrate your own applications
Embedded Linux and containerd support OCI application deployment through the Web Console. Use the Python SDK, REST API and aipc-cli for application integration and management. Event Bus carries structured inference results and device events; MQTT workflows connect applications to an external broker.
Platform services, HAL and the Web Console are available in the NeoRuntime source repository, with application examples and templates for custom development.
Outdoor power and field connections
Power NE503 through an IEEE 802.3at PoE+ switch or a 12 V DC adapter. The camera has IP67 and IK10 protection and an operating range of -40°C to +60°C. Ethernet, alarm I/O and RS-485 support site integration. The enclosure exposes one alarm input.
Applications
Traffic and parking management
Develop vehicle detection and license plate recognition (ANPR / LPR) for parking entrances, access roads and vehicle checkpoints. The Parking Lot showcase publishes vehicle and plate events after the app and its required models are deployed.
Security and perimeter monitoring
Use camera-side person detection at warehouses, industrial sites and restricted entrances. The person-detection example publishes events that custom zone rules can turn into intrusion alerts or local alarm actions.
Industrial safety and PPE monitoring
Evaluate safety helmet detection at construction sites and industrial facilities using a verified application and its separately loaded model. Additional personal protective equipment (PPE) classes require suitable models and application logic.
Visual inspection and equipment OCR
Deploy custom models for product-appearance inspection, label reading or equipment-status OCR. Models, lighting and working distance are selected for the inspection task; application connectors can send results to production or maintenance software.
People counting and occupancy monitoring
Monitor the number of people in defined areas of retail spaces, offices and shared facilities. The verified Occupancy Monitor application publishes occupancy data after app deployment and model loading, providing input for capacity dashboards and site operations.
For an application discussion, contact CamThink Sales with your target objects, working distance and system integration requirements.
Package and setup requirements
Included
- 1 × NE503 camera in the selected configuration
- 1 × mounting screw pack
Customer-provided requirements
- Ethernet cable and either an IEEE 802.3at PoE+ switch, or a 12 V DC adapter with a network switch.
- A computer with Ethernet access and a modern browser for initial setup.
- An MQTT broker and application connection settings when using MQTT event delivery.
A microSD/TF card is optional for additional local storage. Current firmware supports TF expansion; the documented M.2 interface is not enabled. RTSP is the supported video output protocol; RTMP remains planned.
OEM / ODM and project customization
CamThink offers OEM / ODM and project customization for NE503, including lens configurations and communication options tailored to your deployment.
To discuss customization, model compatibility, memory sizing or volume orders, contact CamThink Sales with your project requirements.
NeoEyes NE503 specifications
Models & variants
| SKU | Model / configuration | Memory |
|---|---|---|
| 73.001.000021001 | NE503 · 8-32 mm AF zoom | 4 GB LPDDR4 |
| 73.001.000019000 | NE503 · 8-32 mm AF zoom | 8 GB LPDDR4 |
| 73.001.00002300H | NE503 · 2.8-12 mm motorized zoom | 4 GB LPDDR4 |
| 73.001.00002200G | NE503 · 2.8-12 mm motorized zoom | 8 GB LPDDR4 |
Scroll to view all columns.
Compute and storage
| Item | Specification |
|---|---|
| SoC | Hailo-15H |
| CPU | Quad-core Arm Cortex-A53, 1.3 GHz |
| NPU performance | 20 TOPS (INT8) |
| DSP | 350 GOPS |
| Memory | 4 GB / 8 GB LPDDR4 |
| Internal storage | 64 GB eMMC |
| Storage expansion | microSD/TF, UHS-I; card not included |
| M.2 storage | Not supported by current firmware |
Imaging and optics
| Item | Specification |
|---|---|
| Image sensor | Sony IMX678, 1/1.8-inch CMOS |
| Output resolution | 3840 × 2160 (4K UHD) |
| Image processing | Hailo Gen2 AI-ISP; AI denoising |
| HDR | Digital Overlap / Dual Gain HDR |
| IR-CUT | Electromagnetic switching |
| Image stabilization | Gyroscope-assisted EIS |
| Item | 8-32 mm AF zoom | 2.8-12 mm motorized zoom |
|---|---|---|
| Lens control | Autofocus and motorized zoom | Motorized zoom and focus |
| Optical zoom | 4× | Approximately 4× |
| Maximum aperture | F1.6 | F1.35 (wide) / F2.17 (tele) |
| Horizontal field of view | 44.5° (wide) / 14.5° (tele) | 109.8° (wide) / 46.2° (tele) |
| Vertical field of view | 24.6° (wide) / 8.4° (tele) | 57.2° (wide) / 25.9° (tele) |
Scroll to view all columns.
Video
| Item | Specification |
|---|---|
| Video encoding | H.264 / H.265, hardware-accelerated |
| Maximum encoded video | 3840 × 2160 at 30 fps |
| Rate control | CBR / VBR |
| RTSP streams | Main / sub / third |
| AI frame input | Sub / third raw frames; main stream is encoded only |
Software and AI
| Item | Specification |
|---|---|
| Operating system | Embedded Linux (Yocto) |
| Application runtime | containerd; OCI containers |
| AI inference | On-device, multi-model |
| Pre-installed models | YOLOv8n person detection, Face Landmarks, CLIP Image Encoder |
| Custom model format | Hailo-15H HEF |
| AI output | Structured inference results and events |
| Management | Web Console / SSH / REST API / aipc-cli |
| Application SDK | Python |
Network and external interfaces
| Item | Specification |
|---|---|
| Ethernet | 10/100 Mbps LAN |
| Video transport | RTSP over TCP; RTMP planned |
| ONVIF service | Profile S discovery / stream URI; firmware-dependent, enabled separately |
| Event integration | Event Bus; MQTT via application integration |
| Alarm input | 1 physical input; Event Bus/API reporting unavailable in current firmware |
| Access-control outputs | Wiegand CH0 / CH1, for access-control controllers |
| Serial interface | RS-485 |
| Audio | Line-In / Line-Out |
Power, environment and dimensions
| Item | Specification |
|---|---|
| Power input | IEEE 802.3at PoE+ or 12 V DC |
| Typical power consumption | < 8 W |
| Ingress / impact protection | IP67 / IK10 |
| Operating temperature | -40°C to +60°C |
| Relative humidity | 0 to 95% RH, non-condensing |
| Dimensions | 320 × 134 × 126 mm |
| Weight | 2.5 kg |
NeoEyes NE503 Interface Board Layout

NeoEyes NE503 Main Board Layout

NE503 resources
Evaluate and install
- CamThink's Download Center for NE503 related files
- NE503 product information
- NE503 Quick Start and stream verification
- NE503 installation and connector wiring manual (PDF)
- NE503 hardware specifications
Develop and integrate
- Verified NE503 applications and deployment notes
- NE503 SDK, source code and application resources
- Compile custom models for NE503
- NE503 software architecture
- NeoRuntime SDK source and API clients
For configuration, integration and order questions, contact CamThink Sales.



























