A4K

The drone market is booming, but misuse is rising. We deliver affordable, scalable detection to secure skies and protect privacy.

Comment

Inspiration

The global drone market is expanding at an exponential pace, unlocking new opportunities in delivery, agriculture, and entertainment. Yet, this growth comes with serious risks. Drones are increasingly misused for smuggling, espionage, and terrorism, while unauthorized flights threaten privacy and safety at airports, military bases, prisons, and even private homes. Current detection methods like radar and cameras are costly, infrastructure-heavy, and often inaccessible to smaller organizations or individuals. We saw a clear gap: the world needs affordable, scalable drone detection.

What It Does

Our solution detects unauthorized drones in real time using lightweight technology. Instead of relying on expensive radar systems, we combine acoustic sensing and machine-learning powered audio analysis to identify drone activity quickly and accurately. Alerts are sent instantly to a dashboard or mobile app, empowering security teams and homeowners to respond before threats escalate.

How We Built It

Backend (Python 3.11 + FastAPI)

  • FastAPI handles two interfaces:
    • POST /predict for file uploads
    • WebSocket /ws/audio for live streaming
  • Audio is processed using Librosa + NumPy to generate log-mel spectrograms
  • We run our drone detector model using ONNX Runtime, a fast, cross-platform engine

Frontend (React + TypeScript + Vite)

  • REST and WebSocket clients for backend communication
  • Live prediction display with probability
  • A radar visualization to map recent detections
  • Clean, responsive UI that works in any browser

Model Pipeline

  • Dataset to preprocessing to mel spectrograms
  • Training in PyTorch
  • Exported to ONNX for fast inference

Challenges We Ran Into

  • Differentiating drones sounds from other similar sound sources and background noise
  • Reducing false positives in noisy environments
  • Balancing accuracy of analysis and computational time
  • Ensuring scalability across different environments (urban vs rural)

Accomplishments That We're Proud Of

  • Built a prototype that detects drones without expensive radar or cameras
  • Achieved high accuracy in initial tests
  • Designed a solution that could be deployed universally, in both large facilities and private homes

What We Learned

  • Converting raw waveforms into log-mel spectrogram
  • ONNX for improving deployment by allowing models to run anywhere
  • Designing a user interface that interacts with ML models

What's Next For A4K

  • Improve detection accuracy with larger datasets
  • Mobile app development for ordinary, everyday consumer base
  • API release for commercial use
  • Web socket for live audio analysis
  • LiDAR for precisely determining distance and speed
  • Leveraging pre-built audio models to increase prediction accuracy

Built With

Share this project:

Updates