Abstract
|
|
Public surveillance cameras are increasing in popularity, and as image quality and face recognition software improves, leading to growing privacy concerns of the individuals captured on camera. The ability to collect information on the number of individuals in a public space without identifying anyone would satisfy both those interested in data and those who want privacy protected.
|
Introduction
|
|
Unusual spikes or valleys in populations traversing public spaces has been shown to be an early indictor of major incidents[1]. Such incidents include terrorism and bioterrorism activity. Video surveillance has been proven in previous work to be an effective means of tracking deviations in populations[1].
|
Methods
|
|
Images were automatically obtained from a public webcam hosted by the University of Washington. The images were cropped, as a large portion of the image was of buildings and sky, which were of no interest, and converted to a more suitable image format. Afterwards the images were thresholded against two standard deviations, to eliminate noise, static, and nonhuman features in the image, and the remaining image was scanned for human bodies.
|
Results
|
|
614 useable images were collected for analysis. The final algorithm has an accuracy of .997, a 1-specificity of .001, and a sensitivity of .459, and shows promise for development of larger applications.
|
Privacy
|
|
Counting only bodies a public space can be done without compromising the privacy of the individuals crossing that space. This allows society the benefits of the data without risking the privacy of any of its citizens by protecting all details of their appearance.
|
References
|
|
https://www.washington.edu/cambots/
Project Paper (PDF)
|
