Human-centric computer vision bears many risks concerning the privacy of the depicted users. We investigate alternative sensing modalities, including ultrasound, that natively preserve private information by only providing sparse and uncertain measurements and other security measures such as watermarking datasets and models.