DROPEX

Disaster Rescue Operations and Probing using EXpert Drones is Disaster Risk Management (DRM) framework which is utilised to use swarm of drones with autonomous person detection and payload dropping making the rescue operations optimal.

Architecture

Implementation

As seen in the architecture, the Object Detection model using YoLoV8 and META-DeTr on two different datasets. The model are converted to ONNX format which is sutiable to be run upon embedded systems like Raspberry Pi 4.

These models are deployed using a FastAPI server with streamlit user interface for simulation purposes. The simulation is majorly supported using a WEBGL based app created using Unity Engine and the source code can be found here: DROPEX-simulation.

Running simulation locally:

If you prefer to use your own models, you can replace the models in the simulation/models directory either by providing local path or hosting your model in Hugging Face and providing the model name. The DETR model used by default can be found at Hugging Face: DETR.

cd simulation
python setup.py

Run FastAPI server (also servers the WEBGL app)

cd server
uvicorn main:app --reload

Run Streamlit server

cd client
streamlit run main.py

Note: Before running the servers, add your firebase-adminsdk.json file to both server and client directories. This is required for the firebase authentication. Also, enable read and write operations to be true in the firebase rules for realtime database.