Facial Privacy Protection in Airborne Recreational Videography

Abstract

Cameras mounted on Micro Aerial Vehicles (MAVs) are increasingly used for recreational photography and videography. However, aerial photographs and videographs of public places often contain faces of bystanders thus leading to a perceived or actual violation of privacy. To address this issue, this thesis presents a novel privacy lter that adaptively blurs sensitive image regions and is robust against di erent privacy attacks. In particular, the thesis aims to impede face recognition from airborne cameras and explores the design space to determine when a face in an airborne image is inherently protected, that is when an individual is not recognisable. When individuals are recognisable by facial recognition algorithms, an adaptive ltering mechanism is proposed to lower the face resolution in order to preserve privacy while ensuring a minimum reduction of the delity of the image. Moreover, the lter's parameters are pseudo-randomly changed to make the applied protection robust against di erent privacy attacks. In case of videography, the lter is updated with a motion-dependent temporal smoothing to minimise icker introduced by the pseudo-random switching of the lter's parameters, without compromising on its robustness against di erent privacy attacks. To evaluate the e ciency of the proposed lter, the thesis uses a state-of-the-art face recognition algorithm and synthetically generated face data with 3D geometric image transformations that mimic faces captured from an MAV at di erent heights and pitch angles. For the videography scenario, a small video face data set is rst captured and then the proposed lter is evaluated against di erent privacy attacks and the quality of the resulting video using both objective measures and a subjective test.