Detect. Rescue. Save Lives.
YAQAZ is an intelligent drone-surveillance platform built for civil-defense scenarios. It combines live camera streaming, real-time AI detection, alert reasoning, severity classification, and mission evidence archiving into one operational dashboard.
This repository is currently used as a public project showcase. The implementation code is private for now and may be added later.
This photo was taken during the Innovation and Digital Entrepreneurship Forum for Graduation Projects at the College of Computer Science and Engineering, University of Jeddah.
Senior project poster:
Open YAQAZ Senior Project Poster
YAQAZ is designed to help emergency-response teams monitor a live drone/camera stream and quickly identify incidents that may require action.
The platform currently supports:
- Live camera mission streaming.
- Real-time object detection using custom YOLO models.
- Fire, smoke, and human detection workflows.
- Human-assistance analysis using a vision-language model.
- Fire-severity classification using a dedicated classifier.
- Automatic alert generation with risk levels.
- Mission gallery with saved evidence images.
- Interactive dashboard for mission and alert insights.
- Bounding-box overlays for visual inspection.
- Tailscale/Caddy deployment flow for remote demos.
The platform follows a detection-to-decision pipeline: the camera stream is transmitted, processed by AI models, converted into alert results, stored, and then reviewed through the dashboard.
Camera / Drone Stream
|
v
MediaMTX Stream Bridge
|
v
DeepStream + YOLO Detection
|
v
Decision Layer
- Human scene reasoning
- Fire severity classification
|
v
Backend API + Database
|
v
Frontend Dashboard / Gallery / Alerts
The goal is to keep the live-monitoring experience fast while sending selected evidence frames to deeper analysis only when detection criteria are met.
YAQAZ separates the system into detection, fusion, and decision layers. YOLO models handle real-time RGB/thermal detection, the collected frames and events are stored, and specialized decision models evaluate whether the scene requires action.
The landing page presents the three core areas of the platform: live monitoring, mission gallery, and analytics dashboard.
The live monitor is the operational screen used during a mission. It displays the camera stream, detection boxes, mission state, recent captures, alerts, and live activity updates.
The system can also run fire-focused missions and display detected fire/smoke regions over the live stream.
Alert details include the captured evidence frame, bounding box, confidence, risk level, reasoning summary, observed signals, and recommended action.
The mission gallery stores saved alerts and evidence images. It supports filtering, sorting, selecting records, deleting records, and opening a detailed view for each alert.
The gallery is also used for fire-related incidents and severity review.
The dashboard summarizes mission activity, saved alerts, risk distribution, recent missions, and recent evidence captures. It is intended to help operators understand the current state of the system without reviewing every frame manually.
The dashboard includes mission-level tables and operational summaries for reviewing previous activity.
YAQAZ uses multiple AI layers:
- YOLO detection for real-time object localization.
- Fire detection for fire/smoke regions.
- Human detection for identifying people in the scene.
- Vision-language reasoning for evaluating whether a detected person may need help.
- Fire-severity classification for ranking fire risk.
Alert risk levels are normalized into:
- Normal: no immediate danger.
- Moderate: possible danger or suspicious context that needs review.
- High: urgent or already-happening danger that requires attention.
NVIDIA DeepStream was used to keep video decoding, inference, metadata extraction, and detection throughput close to real-time. This avoids pulling every frame back into a slow CPU-only path and makes the system more suitable for live monitoring.
The prototype demonstrates:
- End-to-end live streaming.
- Real-time detection overlays.
- Alert generation.
- Evidence capture.
- Database-backed gallery.
- Dashboard summaries.
- Remote demo access through a private network setup.
The public repository may contain only documentation and presentation assets while the implementation remains private.
YAQAZ was developed as a graduation project in the civil-defense and intelligent-drone monitoring domain.
The project was showcased at the Innovation and Digital Entrepreneurship Forum for Graduation Projects at the College of Computer Science and Engineering, University of Jeddah.
Project team:
- Naif Almasoud
- Jameel Mahjub
- Ahmed Jamjoom
- Abdulrahman Tibagi
- Abdulaziz Nasrallah
Supervised by Dr. Skander Htiouech.