LAPD
(Laser-Assisted Photography Detection)
(Laser-Assisted Photography Detection)
LAPD aims to detect hidden cameras (even those that do not emit any wireless signals) using only a smartphone. We emit laser signals from the time-of-flight (ToF) sensor on modern smartphones, and use computer vision and machine learning techniques to locate the unique reflections from hidden cameras.
Tiny hidden spy cameras concealed in sensitive locations including hotels and bathrooms are becoming a significant threat worldwide. These hidden cameras are easily purchasable and are extremely difficult to find with the naked eye due to their small form factor. The state-of-the-art solutions that aim to detect these cameras are limited as they require specialized equipment and yield low detection rates. Recent academic works propose to analyze the wireless traffic that hidden cameras generate. These proposals, however, are also limited because they assume wireless video streaming, while only being able to detect the presence of the hidden cameras, and not their locations. To overcome these limitations, we present LAPD, a novel hidden camera detection and localization system that leverages the time-of-flight (ToF) sensor on commodity smartphones. We implement LAPD as a smartphone app that emits laser signals from the ToF sensor, and use computer vision and machine learning techniques to locate the unique reflections from hidden cameras. We evaluate LAPD through comprehensive real-world experiments by recruiting 379 participants and observe that LAPD achieves an 88.9% hidden camera detection rate, while using just the naked eye yields only a 46.0% hidden camera detection rate.
Hidden cameras can be installed into numerous everyday objects and due to the small size of each camera, they are practically impossible to detect with the naked eye alone. The only part of the camera that must be exposed is the lens - a tiny region that is only 1 - 2 mm wide.
In fact, the Bottle and Watch examples on the left are actual commercially-available products that can be purchased for under $20. To test LAPD, we used many other form factors, sizes, and shapes of hidden cameras and everyday objects (e.g., like the Alarm Clock).
Before LAPD, the only approach to detect hidden cameras involved off-the-shelf "hidden camera detectors". A user looking through the detector’s viewfinder observes bright reflections from nearby camera lenses due to the red light emitted from LEDs on the detector. While more effective than the naked eye, for this to be effective, a user must always carry such devices with them. In addition, the detectors exhibit high false positives from reflective surfaces. In the example below, we install a hidden camera into an everyday power plug. When we try to use the "hidden camera detector", the view through it seems confusing. Which of the bright reflective areas corresponds to the hidden camera?
Example of off-the-shelf
"hidden camera detector"
Example of an object with an embedded hidden camera
View through off-the-shelf
"hidden camera detector"
LAPD is implemented as an Android application that uses computer vision, machine learning, and augmented reality techniques to find hidden cameras and display their location to the user.
In this example, the user just completed a scan of an object (the white router) using LAPD. LAPD has detected the hidden camera placed within one of the many holes of the router, and notified the user of its presence with a large green circle over the suspected area.
The time-of-flight (ToF) sensor on modern smartphones uses the travel time of laser emissions to report the distance to nearby objects. It exists for use-cases such as auto-focus and accurate augmented reality. The example below ("ToF: Distances") shows the output from the ToF sensor when pointed at a water bottle.
In LAPD, we use fuse both the distance information and reflected light intensity information from the smartphone ToF sensor to detect areas of unnaturally high reflectivity. These may correspond to the location of hidden cameras.
However we see in the example below ("ToF: Reflected Light Intensity") that only one of the four high-intensity reflection areas (black pixels) correspond to hidden cameras. One of the primary challenges of LAPD is deciding which of these regions (if any) are actually hidden cameras.
LAPD faces several technical challenges as it is intended to be used on existing smartphone hardware, and such hardware is not intended for this purpose. We cover just a selection of these challenges here.
If we stand too close to an object of interest, the ToF sensor may receive reflections that are too strong and overwhelm ("saturate") the sensor. Likewise, if we are too far, we may not receive strong-enough reflections from the object to conclude if there is a hidden camera or not. To solve this problem, LAPD guides the user to the ideal distance using computer vision and augmented reality techniques.
As ToF sensors are not designed for this purpose, they have a relatively low resolution (number of pixels) and low bit-depth (bits per pixel in the reflected light intensity image). Therefore, we face the possibility of numerous false positives as there are many regions of high reflectivity that are hard to differentiate. LAPD removes such false positives using image processing and machine learning filters.
Hidden camera reflections are only visible in a limited region (+/- 10 degrees from the center of the camera) due to the physics of the effect we exploit (retro-reflection). Therefore, it's possible for users to miss hidden cameras if they do not scan the area thoroughly. LAPD guides the user with UI/UX feedback in the app to ensure that they complete a thorough scan of the area.
Publications
LAPD: Hidden Spy Camera Detection using Smartphone Time-of-Flight Sensors, Sensys'21 [ paper ]
On Utilizing Smartphone Time-of-Flight Sensors to Detect Hidden Spy Cameras, Sensys'21 [ poster | poster abstract ]
Code
LAPD source code: https://github.com/frizensami/lapd