SweepLED: AI-Powered Hidden Camera Detection via Smartphone LED

Researchers from KAIST, the National University of Singapore (NUS), and Singapore Management University (SMU) have developed SweepLED, a technology that detects illegal hidden cameras using a low-cost LED attachment and AI-driven reflection analysis. By analyzing how light reflects off surfaces from multiple angles, the system can distinguish between camera lenses and common glossy objects with 94% accuracy in under five seconds.

AI-Driven Reflection Analysis vs. Manual Inspection

SweepLED replaces the manual process of searching for "bright spots" with an automated deep learning approach. Traditional portable detectors require users to shine a light and visually identify reflections, which often leads to false positives from metal, glass, or glossy plastic.

SweepLED solves this by keeping the smartphone camera fixed while the attached LED device continuously changes the direction of the light. Because camera lenses contain a complex internal structure—including multiple lenses, apertures, and image sensors—they produce unique reflection patterns as the light angle shifts. In contrast, reflections from standard glossy surfaces move or disappear linearly. The AI analyzes these time-varying patterns to determine if the reflection originates from a camera lens.

Performance and Hardware Costs

The system is designed for accessibility and low-cost consumer deployment. Key performance metrics include:

Accuracy: 94% detection rate across 30 common household objects (e.g., chargers, alarm clocks, and remote controls).
Speed: Each inspection is completed in less than five seconds.
Cost: The core LED hardware costs approximately $7 (10,000 Korean won).

Technical Alternatives and Limitations

While SweepLED provides a low-cost entry point for privacy protection, technical discussions highlight several alternative detection methods and potential vulnerabilities:

Alternative Detection Methods

Thermal Imaging: Some suggest that handheld thermal cameras are more robust because CMOS sensors, wireless transmitters, and power sources generate heat that can be detected regardless of the lens reflection.
Laser Scanning: High-end detection often involves scanning with a specific wavelength laser (e.g., 832nm) and a mirror to find reflective surfaces.
RF Analysis: Detecting transmission spikes caused by motion in the room can identify wireless cameras that are actively transmitting data.

Potential Vulnerabilities

Technical critics note that certain camera placements can defeat optical detection. Specifically, cameras placed behind fabrics or clothes can identify subjects without being visible to a light-based scan. Additionally, some users suggest that "smart" hidden cameras could be programmed to activate only after a scan has been performed, bypassing real-time detection.

"Cameras can be put in certain angles that it can see with being seen, or how police do it, entirely behind clothes or similar fabric where it can see through and identify but not seen even if you’re looking straight at it."

This research was presented at the ACM MobiSys 2026 international mobile computing conference in June.