Enabling Personal Privacy Protection Preferences in Collaborative Video Observation

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

Enabling Personal Privacy Protection Preferences in Collaborative Video Observation This work devises automated methods to respond to many of the privacy concerns raised as a result of the increasingly pervasive use of video monitoring of public and private spaces, made possible by inexpensive surveillance and storage technology and justified by the need for greater security and vigilance in our society. Furthermore, as a result of continuing progress in the machine understanding of human behavior and event detection through machine learning and computer vision in well specified domains, many beneficial applications of continuous video and sensor monitoring other than physical security have been emerging. However, research that depends heavily upon clinical trials has been impeded by natural reluctance on the part of subjects to participate in studies where a video record of their activity and behavior is created and made accessible to diverse collaborating groups. One example is the continuous monitoring of nursing home patients to enable early detection of changes in behavior and performance that may correspond to environmental or pharmacological changes in their regimen. Others include the study of child development and aberrant behavior in day care, and the monitoring of independently living senior citizens by their families and medical providers. To meet the associated concerns about ill-controlled access to such data even by authorized users, this project develops and evaluates alternative control and access mechanisms for both the capture and display of video and sensor data that enable a layered access to detail and identity in imagery in order to more appropriately reflect how the observations are intended to be used. This is accomplished through means largely enabled by computer vision and computer graphics research, for example, from selectively identifying and obscuring faces, to alternatively substituting wire-frame and stick- figure surrogates, to complete removal with substitution of background data, all without the need for human identification and intervention in the process. The implemented methods recognize and implement the privacy protections provided by law, by institutional policy, and now, most importantly, by the personal preferences of the participating subjects of observation, whether that be at work, at home, or in a hospital. Different groups with differing purposes are allowed to collaboratively use the same recorded data, but each being able only to observe and extract the information that they need and to which they have been selectively permitted. This project enables significant research in social and medical settings, where automated observation and analysis can yield the keys to understanding change and managing interventions. It is important to alleviate the concerns of citizens through trusted mechanisms. Without such capability, the use of continuously captured and automatically analyzed observation data in many fields of scientific, medical and social research, as well as in providing protections for health and safety, will be seriously impeded, if not prevented. The outcomes of this research will ameliorate or overcome these impediments by enabling meaningful access to video observables while respecting the demands of individual privacy in the light of rapidly expanding public and private surveillance. It is anticipated that these mechanisms will enable a much broader comfort level for participation in video- based applications and environments. URL: http://www.caremedia.cs.cmu.edu/