DroneVision SAR - Search and Rescue Detection Platform

Design

https://www.figma.com/design/0M2kLjF1Rnzie143iVOweH/divasai?m=auto&t=p6xFRmAyt8BnfRxb-6

DroneVision SAR is an intelligent search-and-rescue detection system built to support emergency response teams during disaster operations. By analyzing drone camera feeds in real-time, the platform detects people, vehicles, and emergency signals such as SOS gestures - even in low-visibility environments like forests, mountains, and collapsed structures.

The system connects to a React-based command dashboard, delivering instant alerts, GPS coordinates, and visual bounding box overlays to help first responders rapidly locate survivors.

Technologies

Computer Vision & AI

YOLOv8 Object Detection
Custom dataset training (distressed people, vehicles, SOS)
OpenCV, PyTorch

Platform & Integrations

RTSP / Drone Video Stream Processing
GPS coordinate mapping
WebSockets for low-latency alerts

Frontend

React, Material UI
Real-time video overlay visualization
Map integration (Leaflet / Mapbox)

Backend & Deployment

FastAPI / Node.js (stream + detection management APIs)
Docker + GPU acceleration
Edge/Cloud compute compatibility

Features

Real-time drone video analysis with high-precision human detection
Distress state recognition (lying posture, motionless behavior, SOS signals)
Search zone mapping with survivor geolocation pins
Alert system with auto-captured images and confidence scores
Mission dashboard for active rescue tracking
Low-light & thermal compatibility (optional model version)
Offline fallback mode for remote rescue missions

Development

Created a custom-trained YOLOv8 model optimized for aerial angle detection and tiny object resolution
Built a GPU-accelerated detection service capable of 30+ FPS
Engineered an end-to-end streaming pipeline with reduced latency for real-time decision support
Developed a React dashboard to display detection overlays synchronized with drone geolocation
Implemented a modular architecture so additional object classes (e.g., debris, fire hotspots) can be added easily

Challenges

Tiny Object Detection
- Aerial footage often shows humans at very small pixel sizes; required extensive augmentation & tuning
Harsh Environments
- Landscape camouflage and motion blur impacted model accuracy
Latency Constraints
- Optimization work was required to maintain reliable FPS during active rescues
GPS-Vision Alignment
- Converting bounding box location into accurate world coordinates posed mapping challenges

Conclusion

DroneVision SAR demonstrates a high-impact application of computer vision, drones, and mission-critical UX. By enabling faster survivor detection, the platform helps rescue teams act decisively when every second matters.

This project showcases your strengths in:
✔ Real-time AI pipelines
✔ Full-stack engineering for operational technology
✔ Human-centered product design for emergency use cases