Adaptive Mobile Diminished Reality Framework for 3D Visual Privacy

Privacy refers to the right of individuals to keep their personal information and activities confidential, limiting access to others. In the context of images and videos, it is referred to as visual privacy. Visual privacy is more than a personal preference; it is an essential component of human dignity, security, and well-being. Recently, immersive technologies: Augmented Reality/Virtual Reality/Mixed Reality (AR/VR/MR), have gained surging popularity. Current research is focused on making these nascent experiences more engaging while overlooking the privacy concerns raised by having their cameras continually capture the user’s environment. State-of-the-art visual privacy-preserving techniques focus on anonymizing users, or protecting certain types of information, like faces or bystanders, in a 2-dimensional setting where private information is always considered to be in the background. The few works that provide 3D privacy follow the least privilege paradigm, where only the minimal information required to perform a certain task is released. Moreover, previous frameworks struggle to maintain good performance during significant camera motion. We propose a mobile real-time 3D visual privacy-preserving Diminished Reality (DR) system that adapts to motion patterns. The system tracks the user’s movements and modifies the size of the obfuscation region in a Multiplicative Increase Additive Decrease (MIAD) manner. When rapid movements are sensed, the scale of the obfuscated region is increased rapidly (multiplicatively), whereas when the movements are slow, the scale is decreased slowly (additively) to minimize the risk of exposing any part of the private object. Our approach is evaluated during translational and rotational motion patterns, and it significantly reduced the average number of privacy failures from over 18% to less than 1% compared to the state-of-the-art.