How it Works#

The stack ingests an RTSP stream, runs a DLStreamer pipeline that samples frames for VLM inference, and sends results to the dashboard.

System Architecture Diagram

Data Flow#

        flowchart LR
  subgraph FLOW["Data Flow"]
    RTSP["RTSP Source"] --> DPS["DL Streamer Pipeline Server"]
    DPS -->|"1fps AI branch\n(GStreamer gvagenai)"| MQTT["MQTT Broker"]
    DPS -->|"30fps preview"| MTX["mediamtx (WebRTC)"]
    MTX --> DASH["Dashboard"]
    DASH --> METRICS["Dashboard collects metrics\n(CPU, GPU, RAM)"]
  end

System Components#

  • dlstreamer-pipeline-server: Intel DLStreamer Pipeline Server processing RTSP sources with GStreamer pipelines and gvagenai for VLM inference

  • mediamtx: WebRTC/WHIP signaling server for video streaming

  • coturn: TURN server for NAT traversal in WebRTC connections

  • app: Python FastAPI backend serving REST APIs, SSE metadata streams, and WebSocket metrics

  • collector: Intel VIP-PET system metrics collector (CPU, GPU, memory, power)

Learn More#