A Smart Camera That Sends Results, Not Video
USI's AI Smart Camera combines edge inference with PoE, HDMI and MicroSD, and pairs the hardware with a no-code training platform aimed at people who are not ML engineers.
USI has launched an AI Smart Camera integrating a high-performance, low-power edge computing platform with high-resolution imaging and proprietary AI vision software for real-time analysis on the device. Interfaces are Ethernet, Power over Ethernet, HDMI and MicroSD. Resolution, NPU model, TOPS rating and lens options were not disclosed. The camera is already deployed in USI's own manufacturing operations.
Justin Chang, director of the Vertical Mobility Solution Center at USI, said demand for AI-powered vision applications is growing rapidly across manufacturing and that manufacturers are seeking more than camera hardware.
Two of the unglamorous specifications carry the most weight. PoE collapses power and data into one cable, which is why PoE cameras keep winning retrofit installations against higher-specification cameras that need a separate supply — on a brownfield line, the cable route is the project.
Doing inference on the camera is the other. If only results cross the plant network, adding a camera is adding a few hundred bytes per part. If video crosses it, adding fifty cameras is a network redesign, a storage decision and a conversation with IT that will take longer than the vision project itself.
The no-code and low-code training platform addresses the constraint that actually stops factory vision deployments. The shortage is not cameras and it is not compute. It is people who can label data and train a model, and every deployment that depends on a machine learning engineer being available is a deployment that stalls when that person moves to the next site.