Lightweight Intelligent Monitoring Platform Architecture Based on WVP

Modern intelligent monitoring platforms often become bloated due to numerous modules and components, making private‑deployment difficult; therefore a lightweight solution is necessary.

WVP (Web Video Platform) is an out‑of‑the‑box network video platform that implements the GB28181‑2016 standard, handling core signaling and device‑management backend, supporting NAT traversal and brands such as Hikvision, Dahua, and Uniview, as well as standard cascading and forwarding of non‑standard streams.

Based on the open‑source WVP framework, a new intelligent monitoring product is reconstructed and integrated, combining powerful business functions with WVP’s robust video foundation to achieve a lighter, more feature‑complete, and faster‑to‑deploy platform.

The web‑server technical architecture consists of the following components and responsibilities: Camera: provides video streams (GB‑standard cameras must register to the SIP server). SIP Server: handles GB‑camera SIP signaling, registration, heartbeat, stream control, etc. ZLMediaKit: C++11 high‑performance media server offering various streaming protocols. KubeEdge: open‑source edge‑computing platform that pulls streams from MediaKit, extracts frames, invokes AI interfaces, and stores events in Kafka. AI Service: provides facial recognition, intrusion detection, smoking detection, and other algorithms. OSS: stores AI‑generated images and videos uploaded by KubeEdge. Kafka: message queue between KubeEdge and the web server for AI events. Web Server: offers device management, skill management, event management, statistics, and analysis. MongoDB: stores cloud and AI events. MySQL: stores web‑server data. Milvus: vector database for facial feature storage and matching.

The camera‑skill issuance sequence (illustrated in the diagram) involves the front‑end sending JSON‑formatted skill configurations to the web server, which stores them, forwards them to the edge, and processes resulting AI events via Kafka.

The GB‑standard camera SIP registration flow includes pre‑assigning credentials, camera sending a registration request, SIP server responding with a 401 challenge, the camera resending with authentication, and receiving a 200 success response.

The real‑time streaming SIP architecture follows a multi‑step INVITE/200 OK/ACK handshake among the media‑stream sender, SIP server, and media server, culminating in a B2BUA‑mediated media connection that delivers the stream to the receiver.

Key functional highlights of the platform include support for multiple protocols (Hikvision, RTSP, GB), skill configuration and issuance, event alarm and visualization, device management, map positioning, cloud storage playback, real‑time playback, license‑plate library, and face‑library management.

Technical innovations comprise replacing MySQL with MongoDB for high‑volume event storage (reducing write latency from ~90 ms to ~10 ms), leveraging KubeEdge for flexible pod‑level edge scheduling to balance load, and splitting image/video uploads to accelerate alert timeliness.