A Smartphone and a $7 LED Case Can Now Spot Hidden Cameras

It takes about five seconds and a phone case bristling with tiny LEDs to find out whether the smoke detector above a hotel bed is watching back. Researchers at the Korea Advanced Institute of Science and Technology (KAIST), working with colleagues at the National University of Singapore and Singapore Management University, built a system called SweepLED to answer exactly that question, cheaply and without asking anyone to squint at reflections in a darkened room. A hidden camera’s lens bends a sweeping beam of light in a distinctive, repeatable way that an ordinary shiny surface does not, and that gap in behavior is what lets a smartphone paired with a cheap LED case sniff the camera out. Across 30 everyday objects, the system caught the camera hiding among them roughly 94% of the time.

Hidden cameras have become an ordinary hazard of short-term rentals and hotel rooms, tucked into clock radios, phone chargers, and the screws of a bathroom mirror. Existing detectors mostly ask a person to shine a flashlight around and hunt for a telltale bright dot, a method that mistakes glass, chrome and glossy plastic for a lens often enough to be exhausting. The scale of the underlying problem is easy to underestimate: an Airbnb representative testified at a 2024 deposition that the company had generated at least 35,000 customer-support tickets related to surveillance devices over the preceding decade, an investigation by the network CNN found, in a report on that investigation.

How a Sweep of Light Outs a Hidden Lens

SweepLED works by holding the smartphone’s own camera still while an LED-studded case fires light at a target from several angles in quick succession. As the angle of illumination shifts, an ordinary reflective surface, a picture frame, a watch face, a curtain hook, throws back a highlight that simply slides around or disappears. A camera lens is built differently: light bounces off its aperture, its internal glass elements and its sensor before it comes back out, so the reflection distorts and reshapes itself in a way particular to that stack of optics. A deep-learning classifier trained on those reflection sequences, rather than on any single snapshot, does the sorting. Aiming light at a camera lens to deal with it is not a new idea; two decades ago, a Georgia Tech team built the reverse of SweepLED, a tracking system that pointed pulsed light directly at detected camera sensors to blind them rather than simply spotting them.

That sounds tidy in a lab. Real rooms are full of curved, recessed and oddly shiny objects that throw off camera-like glints of their own, and a detector too eager to flag anything reflective would leave a guest checking the minibar and the doorknob along with the smoke alarm.

So the KAIST team designed SweepLED to look at motion, not brightness. Each of the three viewpoints a user sweeps across gets its own light-on, light-off comparison, isolating exactly what changed when the LEDs moved and discarding whatever glare was already sitting in the room. Only regions that keep behaving like a lens, frame after frame and viewpoint after viewpoint, survive to the final classification step.

Jun Han, a KAIST computing professor who led the project and is the paper’s corresponding author, sums up the appeal without much hedging: “Hidden cameras pose a serious threat to personal safety and privacy in everyday spaces.” He adds that the work “combines low-cost smartphone-based hardware with AI analysis to present the possibility of a practical detection technology that even non-experts can use.”

The pipeline runs in four stages. SweepLED first takes the raw video from each viewpoint and subtracts frames captured with the LEDs off from frames captured with them on, canceling out whatever ambient light or fixed glare the room already had going on and leaving only the reflections the sweep itself created. It then boosts local contrast so faint reflective responses show up clearly enough to analyze. A candidate-detection stage flags anything that looks even slightly lens-like, deliberately erring toward false alarms at this point so a real camera never slips through unflagged, then narrows the list to regions that stay reflective across multiple frames and multiple angles rather than flickering once and vanishing. The final classifier, built on a spatiotemporal video model called R(2+1)D, originally developed for recognizing human actions in ordinary video, watches how each surviving candidate’s reflection evolves in both space and time and checks its verdict against all three of the user’s viewpoints before deciding whether a lens is really there.

The Catch in a Five-Second Sweep

The system’s own numbers show the limits of that sweep. Held in an unsteady hand, SweepLED’s accuracy dipped slightly to 93.9%, versus 95.1% when the phone sat still on a tripod, because jitter makes it harder to isolate a clean before-and-after reflection.

That accuracy also comes from a fairly small test: 30 objects, evaluated by the team that built the detector, under a leave-one-object-out protocol rather than a live trial in actual guest rooms. The work is a two-page poster presented at a companion track of a mobile-computing conference, not a full peer-reviewed paper, and the researchers themselves call it a preliminary evaluation rather than a finished product.

None of that makes the underlying idea less useful. A cheap LED case and a phone people already carry is a meaningfully lower bar than a dedicated radio-frequency (RF) scanner or thermal camera, the kind of gear most travelers never pack. If the accuracy holds up outside a research lab, a five-second sweep could become as routine as checking under a hotel bed skirt, something security researchers have been recommending anyway. The bigger test will be whether SweepLED holds its numbers against hidden cameras that were built, or hidden, specifically to defeat a method like this one. SweepLED is not the only academic answer to the problem, either: researchers at Delft University of Technology recently presented a different optical detection system, called NIRF (Neural Infrared Reflectance Field), at the same MobiCom family of conferences, aimed specifically at cameras hidden behind translucent phone and TV screens rather than behind ordinary household objects.

For now, the system exists as a lab prototype and a two-page write-up, not an app in anyone’s pocket. Whether that changes depends on whether a $7 LED case can keep telling a camera lens from a coat hook once it leaves KAIST’s test bench and lands in a stranger’s rental unit.

FAQ