Pradeep Atrey - Security and Privacy

Richardson College for the Environment


Pradeep Atrey Dr. Pradeep Atrey, Associate Professor, Applied Computer Science. Photo By DanHarperPhoto.Com


Big Brother is our Friend

Fulfilling the two opposing mandates is somewhat like walking a tightrope.

Working in homeland security, Dr. Atrey is caught between the conflicting goals of increasing multimedia surveillance capabilities and preserving privacy. Fulfilling the two opposing mandates is somewhat like walking a tightrope. To achieve balance, Dr. Atrey’s answer is to build models and algorithms.

Dr. Atrey has analyzed traditional security methods regarding privacy. Traditional methods focus on automatically detecting and obscuring Regions of Interest (RoI) in the video – for example, a person’s face. However, without 100% accurate RoI-detection methods, privacy is not attained. Furthermore, the hidden or implicit inference channels that can compromise privacy are completely ignored. These channels are location, time, and activity. Dr. Atrey has developed a technology that allows selected functions to be applied to the original video. Prior to viewing for investigative purposes, those channels compromising privacy are blocked, leaving intact the channel required for qualifying the activity as suspicious or normal. For example, obscuring or hiding the name tag on the door will protect the identity of a person who normally accesses that room at a particular time.

As well, accuracy in surveillance control rooms is at risk. An operator usually monitors up to 16 cameras at a time. Given that a human being can effectively monitor only four camera views  simultaneously, improving accuracy lies in reducing the number of screens. Dr. Atrey has developed an algorithm dubbed Dynamic Selection and Scheduling of CCTV Views. Computers are programmed to automatically process the information of the 16 cameras and to transfer views of suspicious activity to four enlarged screens. Human confirmation as to the relevance of the assigned importance to these views is essential. Computer observation of the operator’s eye orientation provides this feedback. If attention is fixed elsewhere than on the selected views, the top four screens change views accordingly. Operator detection  accuracy has been increased from 70% to an impressive 85%.

Some monitoring activities require multiple cameras in close proximity. Dr. Atrey has programmed the computers to fuse the multi-camera data and work both in competition and in cooperation. Recording a suspicious activity can involve multiple tasks. For example, in the event of a person abandoning a suspicious object in the hallway, there could be three tasks. In order of priority, they could be identification of the person, identification of the abandoned object, and tracking of that person. Initially all three cameras compete for person identification. When that task is better accomplished by a particular camera, the other two compete for the second priority task. Preliminary controlled tests have proven to be extremely successful and indicate worthwhile results from field testing.

Big brother isn’t so scary after all.