AI Video Analytics Edge Box — Operating Guide
How to power on the AI video analytics edge box, put it on the LAN, add cameras, configure algorithms and analysis tasks, and review alarms.
ReadZHICLOUD
Zhcloudlink articles on edge computing, vision models, and AI chips.
These are newly published technical articles. Existing analyses remain on the technology page. View technology articles
How to power on the AI video analytics edge box, put it on the LAN, add cameras, configure algorithms and analysis tasks, and review alarms.
ReadAlarm-push API for the AI video analytics box. A third-party platform receives detection results at a callback URL it configures.
ReadUltralytics has not released YOLO27 yet. This article compiles the official preview: COCO mAP 42.3 to 60.4, and compares accuracy, parameter count, and architecture with YOLO26 and YOLOv8, including how to choose N/S versus M/L.
Read
A GPU (graphics processing unit) and an NPU (neural processing unit) are different kinds of processors. Both speed up computation, but they differ in design goals, architecture, and the jobs they are built for.
ReadA comparison of process node, CPU, NPU, GPU, memory, and video capability for Rockchip RK3588, RK3576, and RK3572, and how the three differ when selecting a platform for a vision agent, multimodal dialogue, and a complex task agent.
Read
A Zhcloudlink technical article on how an RK3572 host plus an RK1828 coprocessor splits an on-device agent workload. RK1828 is 20 TOPS, with 5GB of on-chip DRAM and bandwidth up to 1 TB/s. A 7B model at W4A16 is about 59 tokens/s.
Read