Stealth vision for protecting privacy

We propose an anonymous video capturing system, called "stealth vision", which protects the privacy of objects by fading out their appearance. In order to avoid difficulty in detecting the region of human faces (privacy-protected regions) in an image captured with a mobile camera, the fade-out areas are determined in 3D (three-dimensional) space instead of 2D image space. The 3D position of the captured object is estimated with a simple algorithm, which projects the foreground region in the overhead image onto several horizontal 3D planes and extracts the 3D position while merging the projecting results. By projecting the 3D information onto an image plane of mobile camera, the region to use for the anonymizing process is determined. Experimental results show that our system successfully protects the privacy of target objects.