Published: 2021-07-02 | English reference version of the Chinese source article.

Where the product range is used

XINHUO AI Edge Box products are designed for local video analytics projects in fuel stations, factories, power sites, chemical sites, construction areas, parks, communities and warehouses. They can take video from cameras and NVRs, run local AI tasks and deliver events to an existing platform.

The product range includes 2, 8, 16 and 32-channel configurations. Channel count is a planning reference. Final capacity depends on stream resolution, codec, model type, number of tasks on a stream, analysis frame rate, event evidence and platform delivery load.

Software functions that accompany the hardware

The delivered platform covers device onboarding, live preview, algorithm binding, region drawing, event records, alarm review, push settings and system status. RTSP, ONVIF and GB28181 access can be used where the existing system exposes the appropriate stream or interface.

HTTP and MQTT integration can deliver structured event data to customer platforms. The current public English Event Message API Reference V2.3.2 describes the main fields, evidence options, algorithm identifiers and delivery practices for external systems.

Algorithms and site rules

A delivered project can use PPE tasks, behaviour rules, smoke and flame, intrusion, vehicle and plate events, equipment and route conditions, camera state monitoring, and custom models where the scene needs them. Each task needs a suitable view and a rule that defines the region, time, duration and alarm interval.

The box can be part of a wider solution that also uses AI cameras for new points, AI NVR for recording and review, a video analytics server for larger channel counts, and XINHUOAI for a new model. The platform receives the events from those devices under the customer's integration plan.

What a project team should provide before selection

Useful project material includes a camera and NVR inventory, sample streams or recordings, target algorithms, expected channel count, network topology, required event fields and the platform interface. This turns a hardware quotation into a deployable design.

A completed project should keep its camera record, rule settings, model version, event examples and acceptance results. They are the reference for later maintenance and for a future extension to another point or site.