Using the time-of-flight (ToF) sensor on commodity smartphones, researchers at the National University of Singapore and Yonsei University have developed a technique to detect tiny hidden spy cameras concealed in privacy-invading locations, such as hotels and bathrooms. With small form factors and lens diameters as small as 1 to 2 millimeters, hidden cameras are difficult to detect — even though they are hiding in plain sight.
Called laser-assisted photography detection (LAPD), the technique is implemented as a smartphone app that emits laser signals from the ToF sensor and uses computer vision and machine learning to locate the unique reflections from hidden cameras. The ToF sensor technique is an alternative to wireless detection approaches that other researchers are pursuing by detecting wireless traffic. LAPD does more than detect the presence of the camera. It also locates the hidden camera that may only record the data on a local memory card and not transmit any wireless data.
Challenges that had be overcome include:
- The varying reflectivity of objects that embed the hidden cameras
- Reliably (minimal false positives) identify hidden camera (constrained by the ToF sensor’s number of pixels and number of bits per pixel)
- Limitation of reflections due to their optical properties
To address the first issue, the ToF sensor in the phone must be located an ideal range from the hidden camera. Using augmented reality, LAPD guides the user to the ideal location. The ideal distance is calculated by determining the objects reflectivity at various distances.
To overcome the second challenge, researchers designed and implemented a chain of filters including deep-learning based filters that incorporate multi-modal information including depth and reflection intensities. This eliminated false positives.
Third, since the reflections are limited by their optical properties, they are only observed within a constrained angle of a 20° field-of-view (FoV) cone. The solution was an FoV filter to eliminate the remaining reflections – or potential hidden cameras. These appear highly reflective outside of the constrained angle.
How effective is this approach? Using 379 observers, LAPD found 88.9% of the hidden test cameras vs. 46% discovered by the human observers.