Sec-01 Smart Video Management Platform for Real-Time Event Detection PFE

Objective

• Design and prototype a Video Management System (VMS) module that uses AI to detect specific events (e.g. intrusion, abandoned object, crowding) from IP cameras in real time.
• Produce a working prototype capable of ingesting RTSP streams, performing object detection/event classification, and generating real-time alerts (dashboard + notifications). Optional: integrate the module into an existing open-source VMS (ZoneMinder or Shinobi).

Key components & tasks

• Video ingestion from RTSP streams using tools such as GStreamer or FFmpeg; ensure stable, low-latency stream handling and reconnection logic.
• Object detection and event classification using AI models (YOLOv8, MobileNet, MediaPipe, TensorFlow/PyTorch); implement detection pipeline and event rules (e.g., intrusion zones, abandoned object timeout, crowding thresholds).
• Real-time alerting system using Redis and WebSocket for push notifications, plus a dashboard UI (Streamlit or React) for visual monitoring and alert management.
• Optional task: integrate prototype into an open-source VMS (ZoneMinder or Shinobi) to demonstrate interoperability (APIs, stream linking, event forwarding).

Technological environment / tools

• Languages: Python (OpenCV, TensorFlow/PyTorch) for vision pipeline; Node.js for backend components if needed.
• Tools: GStreamer / FFmpeg for stream handling; YOLOv8 or MediaPipe for detection; Redis + WebSocket for real-time push; Streamlit or React for front-end.
• Deployment: containerize with Docker and target local server or edge devices (e.g., NVIDIA Jetson) for on-device inference and low-latency operation.

Deliverables & evaluation criteria

• A dockerized prototype demonstrating real-time detection on live RTSP streams, with dashboard showing video, detections, and alert logs.
• Implementation of at least two event types (e.g., intrusion and abandoned object) with measurable latency and accuracy metrics; documentation of model choice, training/fine-tuning (if any), and trade-offs.
• Clear README / deployment guide for local or edge deployment (including NVIDIA Jetson notes) and, if performed, integration notes for an open-source VMS.

How to apply

• Apply via the project link: https://www.hydatis.com and/or by email to stages@hydatis.fr.
• Use the email subject: "Application Sec-01 - Smart Video Management Platform for Real-Time Event Detection" when applying by email.