SwiftTrack+: Fine-Grained and Robust Fast Hand Motion Tracking Using Acoustic Signal

作者：Yongzhao Zhang, Hao Pan, Dian Ding, Yue Pan, Yi‐Chao Chen, Lili Qiu, Guangtao Xue, Ting Chen, Xiaosong Zhang · 发表于：IEEE Transactions on Networking · 年份：2024 · DOI：10.1109/tnet.2024.3504517 · 被引用次数：4 · 研究领域：Muscle activation and electromyography studies、Hand Gesture Recognition Systems、Shoulder Injury and Treatment

Acoustic tracking technology, leveraging the ubiquitous presence of speakers and microphones in commercial off-the-shelf (COTS) mobile devices, has become a versatile tool across various applications. However, current phase-based acoustic tracking methods encounter significant limitations in tracking fast movements, thereby restricting their practical utility. This paper identifies three practical challenges to enable fast hand motion tracking using acoustic signals: 1) high mobility, 2) low signal-to-noise ratio (SNR), and 3) variations in hardware frequency response. The high mobility introduces Doppler shift and phase ambiguity which is the primary cause of failure in fast movement tracking, while the latter two factors can further impair the tracking performance in practical scenarios involving high mobility. To address the high mobility issue, we effectively compensate the Doppler shift in the Channel Impulse Response (CIR) for better selection of channel taps and then propose a novel phase derivative approach to mitigate the phase ambiguity. To enhance the real-world robustness, we integrate multiple algorithms including an SNR enhancement algorithm inspired by time-domain beamforming and a hardware frequency response compensation approach that addresses both amplitude and phase distortions. Additionally, an LSTM-based distance reconstruction algorithm is further implemented to correct residual phase noise. Implemented on Android platforms under the name SwiftTrack+, ou...