• about 1 min read

Port Intelligent Perception & Management Platform

An end-to-end multi-camera visual localization platform for a container terminal in Shenzhen, serving autonomous-vehicle dispatch. Four Go/Python services plus a React console, with a strict split between the data plane (ZeroMQ video frames) and the control plane (Kafka 4.x KRaft): FFmpeg-based source management, YOLO detection with a terminal-specific model (7000+ frames labeled in self-hosted CVAT), cross-camera multi-object tracking via a “region + territory” design with global ID management over world coordinates, homography-based pixel→world mapping into Redis, a Gin HTTP API server, and a pluggable-runtime orchestrator driving container lifecycles. Verified end-to-end on real terminal footage. Due to an NDA this page only sketches the architecture — details in the Chinese version (architecture only; no project-specific assets).

jcbl1
jcbl1 Software Engineer