Propeller signatures enable reliable drone tracking from a moving event camera

Propeller-Based Drone Tracking with a Moving Neuromorphic Camera (ICONS 2025) addresses one of the fundamental challenges in event-based drone tracking: separating the event signatures generated by the drone from the event signatures generated by camera motion. When the camera is moving, the background itself generates events, and a naive detector cannot distinguish background motion events from target motion events.

The propeller signature as a feature

The paper’s central insight is that drone propellers generate a highly distinctive event signature: rotating at 4,000–8,000 RPM, the blade tips produce periodic, high-rate events that repeat at frequencies far above what static scene features or typical camera motion generates. A 5-inch propeller spinning at 6,000 RPM produces blade-tip events at a rate exceeding 200 Hz per blade, with a spatial pattern that rotates predictably.

This signature is essentially unique to rotary-wing aircraft. No common background feature — trees, flags, people — generates events at the same rates and with the same rotational periodicity.

The paper exploits this by filtering the event stream for frequency bands that correspond to propeller rotation rates, producing a masked event stream that contains almost exclusively drone-generated events even when the camera itself is moving aggressively.

The moving camera challenge

Camera ego-motion compensation is a fundamental problem in event-based vision. When a camera moves through a scene, every static surface generates events at a rate proportional to camera angular velocity. For a camera mounted on a moving platform — a person, a vehicle, or an airborne interceptor — the background event rate can dominate the total event stream, drowning out the target.

The standard approach to ego-motion compensation is to model the camera’s angular velocity (from an IMU) and subtract the predicted background events from the measured event stream. This works well when the background is static, but in practice the background contains moving elements — trees in wind, passing vehicles — that cannot be subtracted with a pure ego-motion model.

The propeller-signature approach sidesteps this by looking for features that the background cannot produce, rather than trying to subtract the background explicitly.

Performance on a moving gimbal platform

The paper evaluates on a gimbal that moves the event camera through typical hand-held and small-vehicle motion profiles while a drone flies within the field of view. Detection and tracking performance is maintained at camera angular rates up to 150 deg/s — significantly faster than typical hand-held camera motion.

For a seeker application, camera angular rates can reach several hundred deg/s during the terminal phase as the seeker head tracks the target. Whether the propeller-signature approach holds at these rates is not evaluated, but the principle — use a target-specific feature rather than background subtraction — is applicable to any high-motion scenario.

Extension to non-propeller signatures

The paper briefly notes that the method generalizes to any target that produces periodic, spectrally distinctive events — which includes some jet aircraft (compressor blade signatures in the MWIR) and could potentially include other target classes with periodic motion elements.

For fixed-wing UAS without visible propellers, this approach is not directly applicable. But for the Group 1–3 multirotor targets that constitute the most immediate C-UAS threat, it provides a sensor-level signature that goes beyond position and into target classification.

---

Propeller-Based Drone Tracking with a Moving Neuromorphic Camera. ICONS 2025.