Objective

• Design and prototype a VMS module that uses AI to detect specific events (e.g. intrusion, abandoned object, crowding) from IP cameras in real time.
• Deliver a working prototype demonstrating real-time detection, alerting, and a user interface for monitoring events.

Key components / Features

• Video ingestion from RTSP streams with reliable frame handling and low latency.
• Object detection and event classification using AI models (examples: YOLOv8, MobileNet) and logic to translate detections into events (intrusion, abandoned objects, crowding).
• Real-time alerting system including a dashboard and notifications (push via WebSocket/Redis-based pipeline).
• Optional integration with an existing open-source VMS such as ZoneMinder or Shinobi for full-system compatibility.

Technological environment / Tools

• Languages and libraries: Python (OpenCV, TensorFlow or PyTorch), Node.js for backend if needed.
• Video stack: GStreamer or FFmpeg for stream handling and ingestion.
• Detection frameworks: YOLOv8 or MediaPipe for model inference and tracking.
• Real-time components: Redis for pub/sub/cache and WebSocket for push updates to clients.
• Front-end options: Streamlit or React to build the monitoring dashboard and alert UI.
• Deployment: Containerize with Docker; target local server or edge device (e.g. NVIDIA Jetson) for on-premise or edge inference.

Deliverables & evaluation

• A prototype VMS module capable of ingesting RTSP streams and running detection in real time with demonstrable latency and accuracy metrics.
• A dashboard UI showing live video, detected objects/events, and an alert/notification panel.
• Documentation including setup instructions (Docker), model choice and training/inference considerations, and integration notes for ZoneMinder/Shinobi if implemented.

How to apply

• To apply, send your application specifically for this project to: stages@hydatis.fr.
• You may also consult the company site: https://www.hydatis.com for more information about the team and technologies.