Respectful Cameras for Security with Privacy

An emerging class of digital video cameras provides unprecedented ability to zoom in and capture high-resolution video images. This capability is desirable in many applications from security to public relations.

However, such high-resolution scrutiny raises significant privacy concerns. CITRIS researchers Ken Goldberg and Deirdre Mulligan are investigating a new approach to providing a measure of visual privacy by masking an individual’s identity while allowing observation of his or her physical actions and other motion in the scene. Their objective is to develop “respectful cameras.” The key idea is to introduce wearable “markers” that can be detected by image processing software in real time.  For example, the software can track markers, such as inexpensive hats of a particular color or pattern, at the border of the space where the camera is present.

The researchers have created a Respectful Cameras visual privacy system that tracks visual markers to robustly infer the location of individuals wishing to remain anonymous. The resulting image is that of people with colored elliptical disks obscuring their face, while allowing their bodies, motions, and actions to be observed.  The researchers use a static-image classifier that determines a marker’s location using pixel colors and an AdaBoost statistical classifier. They then extended this to marker tracking and a marker model which incorporates velocity and inter-frame information by using a particle filter based approach.

Experiments are currently underway at the CITRIS construction site to evaluate the perceptions of construction workers regarding being observed by a robotic camera while they work.

In future work, experiments will be undertaken with different markers to identify preferred colors or patterns. It may be possible to build a Respectful Cameras method directly into the camera (akin to the V-chip) so that faces are encrypted at the hardware level and can be decrypted only if a search warrant is obtained.

More information, sample videos, and press coverage:

An example of our system obscuring the face of the construction worker wearing a green vest.

An example of the power of our an off-the shelf camera (Panasonic HCM-280) that costs less than $1000.  The left image shows the view from the camera completely unzoomed.  The red square in the center is the barely visible parking sign.  The right image shows the camera at maximum zoom, where even the parking times are legible.