The story

This project explores an edge AI based approach to drone detection using sensor fusion between audio and visual data. The motivation behind this project comes from the limitations of existing drone detection systems that rely almost exclusively on computer vision, which can fail in low light conditions, long range scenarios, or visually cluttered environments.

Drones generate distinctive acoustic patterns due to their propellers and motors, making sound a valuable signal for early detection. This project uses a microphone to continuously monitor the environment and applies an on device machine learning model to classify drone related audio patterns in real time. Audio based detection allows the system to identify potential drones even before they become visible to a camera.

Once an audio event is classified as a potential drone, the system uses visual input to confirm the detection. A camera captures visual data that is processed on the edge device to verify the presence of a drone and reduce false positives caused by environmental noise or other sound sources. The final decision is produced by combining audio and visual information directly on the device through sensor fusion logic.

The entire system is designed to run locally on an Arduino UNO Q using Edge Impulse for data collection, model training, and deployment. The project is intentionally developed in stages, starting with audio only classification and progressively evolving toward full audio and visual sensor fusion. This approach allows for incremental validation of each sensing modality while demonstrating the real benefits of multi sensor fusion at the edge.

At this stage, the project is in the proposal and design phase. Hardware integration, dataset collection, and model training will begin upon selection in the Sensor Fusion Challenge.