Application-Oriented Privacy Filter for mmWave Radar
Hankai Liu, Xiulong Liu, Xin Xie, Xinyu Tong, Tuo Shi, Keqiu Li
Abstract
The high spatial resolution and high displacement sensitivity of mmWave radar have significantly enhanced the performance of wireless sensing. However, improved sensing capabilities have also led to growing concerns about privacy leakage. Attackers can make fine-grained analysis about various user activities by exploiting radar data. This article presents mmFilter, a privacy-level radar data filter designed to protect user privacy, which proposes the signal reversion methodology to perform targeted perturbations in specific feature dimensions, and reverses the altered data back to the raw format for data uploading and subsequent sensing processing. Various perturbation techniques are designed for different sensing functions, which gives mmFilter an application-oriented property. A key design guideline of the mmFilter is that the perturbation should be minimized so that each sensing function can be disabled without affecting the others. In-depth studies on radar data composition, sensing procedures, and human activity features ensure the robustness and completeness of the privacy filter, which are demonstrated by detailed experiments.