Effect of Face Obfuscation Methods on Pose-Based Action Recognition

Vision-based at-home health monitoring is a growing and effective modality for unobtrusively and inexpensively tracking the health of vulnerable people without the kind of sacrifice of autonomy that accompanies residing in a hospital or a care home. The vision modality comes with its own set of unique challenges, principally the preservation of privacy. This paper demonstrates, using three face obfuscation methods, that pose-based actionrecognition is essentially unaffected by privacy-preserving routines.