SUPERSCI-Z

Moving light helps a phone locate hidden cameras

The SweepLED prototype combines an LED phone case with AI analysis; 94% in testing is promising, but it is not yet a guarantee of real-world protection.

Esquema do SweepLED mostrando um celular com uma capa de LEDs que ilumina objetos por diferentes ângulos para identificar lentes ocultas.
Image: KAIST / equipe SweepLED
SUPER SCI-Z editorial analysis

Observation — SweepLED keeps the smartphone camera still while a case switches LEDs on in sequence, illuminating the same object from different angles. A short video records how each bright spot moves, fades, or deforms; a machine-learning model searches that changing pattern for the optical signature of a concealed lens.

Observation — The researchers' experimental set contained 30 everyday objects: 12 held hidden cameras and 18 were reflective objects without cameras. The system reached 93.9% overall accuracy, detected 95% of the cameras, and produced a 7.2% false-positive rate on ordinary objects; each scan took under five seconds, and the case's core components cost less than US$7.

Inference — The improvement does not come from AI magically recognizing a camera, but from a more informative optical measurement. On shiny metal, glass, or plastic, a reflection tends to shift or disappear as the light source changes; inside a lens, optical elements, an aperture, and a sensor reshape the reflected pattern in a characteristic way.

Hypothesis — Separating the viewing position from the direction of illumination should make lenses easier to distinguish from false alarms than searching for one bright point. The results support that hypothesis within the tested set, including handheld use, but they do not yet show that performance will carry over to every room, device, distance, coating, or deliberate concealment strategy.

Practical limit — Thirty objects make a proof of concept, not a safety certification. Aggregate accuracy does not remove errors: a camera can be missed and an innocent object can be flagged. Commercial use would require larger independent datasets, comparisons with existing detectors, and testing across phone models, lighting, distance, and user experience.

Responsible speculation — A low-cost case might make hotel or rental inspections more accessible, or become a screening tool for security teams. That is an engineering possibility, not a proven product or evidence of crime prevention; any alert should be documented and reported to the venue or authorities without creating a risky confrontation.

03

Key points

  • SweepLED uses changing light angles and reflection dynamics to separate lenses from shiny surfaces.
  • Its 93.9% accuracy came from only 30 objects and does not guarantee everyday performance.
  • This is a promising privacy prototype, not an infallible commercial detector.
Primary sourceKAIST Breakthroughs