Space target detection with event cameras demonstrates low-latency detection against point sources in complex backgrounds

Space surveillance and C-UAS have more in common technically than the mission context suggests. In both cases, the object of interest is small, fast, and low-signature against a structured background. In the space case the background is star fields; in the UAS case it’s sky clutter and terrain. In both cases the target produces a characteristic motion signature — a streak or a point that moves consistently while the background is approximately fixed — and the detection problem is finding that motion signature in a noisy background.

A Springer 2025 paper (from the broader research program that also includes the Optica COTS event camera work we’ve followed) applies event cameras to space target detection. The basic hypothesis is that an event camera, pointed at a star field with a satellite in view, will generate events primarily from the satellite as it moves across the background stars. The stars are stationary (relative to the telescope, which is tracking), so they produce no events. The satellite, which moves against the star field, triggers events at every pixel it crosses. The result is a very clean detection signal with minimal clutter.

The experimental setup

The paper uses a ground-based telescope with an event camera mounted at the focal plane. The telescope tracks at sidereal rate, so stars appear stationary in the event camera frame and the satellite appears to move. Targets range from low Earth orbit CubeSats — roughly one arc-second angular size at 500 km altitude — to decaying debris objects that produce longer event trails as they pass through the field of view.

Detection rates on the benchmark are high: better than 98% probability of detection for LEO objects above a minimum brightness threshold, with a false alarm rate far below what comparable frame-camera systems achieve. The latency advantage is also significant — the event camera produces a detectable event train within 50 microseconds of the satellite entering the field of view, while a frame camera with a 100 ms exposure would detect it at best 100 ms later.

How this generalizes to air targets

The sky-clutter scenario in C-UAS is harder than the star-field scenario in space surveillance, because the sky has structure — clouds, terrain features at low elevation angles, aircraft contrails — that produces background events. The space surveillance result works partly because the background is genuinely uniform in the relevant time scale. The sky is not.

But the key principle transfers: a small target moving consistently against a background that doesn’t move as fast produces a distinctive event signature. Background clutter events are random in direction; target events are correlated in direction because the target follows a trajectory. A simple velocity filter — retaining only events whose local motion direction is consistent with a plausible target trajectory — rejects a large fraction of clutter events while retaining the target events.

This is the Hough-based or accumulation-based detection approach that several papers on event-camera UAS detection have explored. The space surveillance paper provides evidence that the underlying detection principle works at the sensitivity levels needed for very small, dim targets, which supports confidence that air target detection with somewhat larger signature margins is achievable.

Latency at the limit

The 50-microsecond detection latency for a space target is as close to the physical limit as you can get — it’s essentially the propagation time from the sensor pixel to the processor. No frame-based system can achieve this. For terminal C-UAS, the time constant of interest is different — you’re not detecting a new target, you’re maintaining tracking of a known target through a terminal maneuver — but the principle is the same: the event camera can provide track updates on microsecond timescales, which is what the guidance law needs to remain accurate through evasive maneuvers.

The space surveillance work is a useful proof point that event-camera sensing works at the extremes of the detection problem. If it works for a one-arc-second CubeSat against a star field with 50-microsecond latency, the equivalent problem for a Group 1 drone at 200 meters with a 1 ms latency requirement is firmly within the capability envelope.