Privacy-Preserving Human Behavior Estimation with Sparse Sensor Measurements

Abstract

Most conventional methods for person behavior estimation acquire information that can easily identify the person, such as images that include information on the face and clothing or acoustic information that includes voice, speech content, musical instrument playing sound, etc. Instead, we have been tack-ling proposed methods that do not acquire them towards privacy-preserving human behavior estimation. Specifically, we have been working on 3D human pose estimation with binary images [1], laser reflection [2], acoustic signals [3], and human mesh estimation with event data [4], making it possible to estimate human behavior in various situations while protecting personal information. In this talk, I will introduce our recent work on these topics.