mmWave FMCW Radar-Based Detection and Tracking of Rotary-Wing Drones

Sammanfattning

The increasing availability of small commercial drones poses a growing challenge to the security of critical infrastructure and restricted areas. This project investigates the use of a millimeter-wave (mmWave) FMCW radar for detecting and classifying small rotary-wing drones at short range using affordable, portable hardware. A prototype system was developed using the Texas Instruments AWR1843BOOST radar sensor and a Raspberry Pi 5. Since the radar’s built-in CFAR detector suppresses zero-Doppler returns, a custom range profile detector was developed as the primary detection method for hovering drones. It operates on pre-Doppler data and tracks persistent energy peaks across frames. Classification is performed through micro-Doppler analysis using spectral features from rotating propellers. The radar firmware was modified to extract complex I/Q samples from the radar cube over UART, eliminating the need for external ADC capture hardware. A pan-tilt servo points towards confirmed detections. The system was evaluated through ten outdoor experiments using a DJI Mavic 2. The range profile detector achieved 100% detection at 2 m, 85.9% at 4 m, and 32.8% at 6 m. Micro-Doppler classification achieved 79.9–91.3% recall with 100% precision for drone-only scenarios. However, the classifier produced a 63.0% false positive rate against a walking person. Aggregated results show a precision of 81.7%, recall of 65.6%, and F1-score of 72.7%. The results demonstrate thatmmWave FMCW radar can detect hovering drones at short range with low-cost hardware, but threshold-based Classification is insufficient for separating drones from human targets. Future work should focus on machine learning-based classification.