Industrial cameras have traditionally done one job: capture images and send them somewhere else for processing. Whether that destination is a local server or the cloud, the camera has largely been responsible for collecting data rather than interpreting it. This solution works; however, it also introduces latency. When a camera is responsible for inspecting products on a fast-moving assembly line, protecting worker safety, or keeping a facility secure, those delays may be unacceptable.
A new edge AI camera platform called NeoEyes NE503, recently launched by CamThink, was designed to change this paradigm. NeoEyes NE503 doesn’t just stream images to a remote server for processing — it does the job itself. The camera is equipped with all the hardware needed to provide 20 TOPS of on-device inference.
The camera handles image processing onboard (📷: CamThink)
The platform is built around the Hailo-15H processor, which combines a quad-core Arm Cortex-A53 CPU with dedicated AI acceleration. Thanks to its 8 GB of LPDDR4 memory and 64 GB of eMMC storage, the system is capable of running multiple vision models locally while simultaneously recording 4K video. Rather than relying on external hardware, developers can deploy complete computer vision applications directly on the camera using containerized software.
Images are acquired by a Sony IMX678 1/1.8-inch CMOS sensor capable of capturing 4K video at up to 60 frames per second. The camera also includes HDR support, an AI-assisted image signal processor, autofocus, and a motorized 8-32 mm lens with 4x optical zoom. Together, these features help produce high-quality images even in challenging lighting conditions while improving the accuracy of AI-based recognition tasks.
Instead of being limited to a single purpose, NeoEyes NE503 is intended to serve as a programmable vision platform. CamThink provides an embedded Linux environment built with the Yocto Project, along with a Python SDK, REST API, SSH access, and support for Python, Go, and C++ development. Applications are packaged as OCI-compatible containers, allowing developers to update or replace individual services without reinstalling the entire system.
NE503 is ideal for a wide range of applications (📷: CamThink)
The camera supports a wide range of computer vision workloads, including object detection, optical character recognition, face detection and recognition, person re-identification, pose estimation, behavior analysis, and attribute recognition. Several pretrained models, including YOLOv8n for person detection, are already installed to simplify development and testing.
NeoEyes NE503 was also designed to fit into existing industrial and security systems. It supports RTSP video streaming alongside structured event output through REST APIs and an internal event bus. Physical interfaces, including alarm inputs and outputs, RS-485, and Power over Ethernet make it possible to connect directly to access control systems, warning devices, sensors, and industrial equipment without additional gateway hardware.
The IP67-rated enclosure allows the camera to operate in temperatures ranging from -40°C to 60°C, making it suitable for both indoor and outdoor deployments. CamThink expects initial shipments of the $1,199 NeoEyes NE503 to begin in late July 2026, with the platform targeting system integrators, OEMs, and developers building custom AI-powered vision solutions.