PiSAR: Pipeline for Aerial Search and Rescue AI-powered visual detection pipeline for aerial search operations

Try PiSAR online:

Installation

Prerequisites

• Python 3.12+
• pip3 (Python package installer)
• (Optional) CUDA-enabled GPU

Setup

1. Clone the repository

   git clone https://github.com/eadali/PiSAR.git
   cd PiSAR

2. (Recommended) Create a virtual environment

   python3 -m venv pisar
   source pisar/bin/activate

3. Install dependencies

   • CPU only:

     pip3 install -r requirements.txt

   • GPU (CUDA) support:

     pip3 install -r requirements-cuda.txt

4. Verify installation

   python3 -c "import torch; print(torch.cuda.is_available())"

See requirements.txt and requirements-cuda.txt for details.

---

Usage

Running the Script

To process an image, video, or camera stream, use the following commands:

Process a video (CPU)

python3 demo.py config/yolo8n-bytetrack-cpu.yaml --onnx-path downloads/yolo8n-416.onnx --video downloads/forest.mp4

Process a video (CUDA/GPU)

python3 demo.py config/yolo8n-bytetrack-cuda.yaml --onnx-path downloads/yolo8n-416.onnx --video downloads/forest.mp4

Command-Line Arguments

Argument	Description	Required/Default
config	Path to the YAML configuration file	Required
--onnx-path	Path to the ONNX model file	Required
--image	Path to the input image file	Mutually exclusive with --video/--camid
--video	Path to the input video file	Mutually exclusive with --image/--camid
--camid	Camera ID for video capture	Mutually exclusive with --image/--video

Note:

• You must provide exactly one of --image, --video, or --camid.
• The config argument is a positional argument (no --config).

---

Acknowledgements

Special thanks to the StephanST for providing the WALDO30 model used in this project.