基于稀疏傅里叶变换和纠错码的实时无损无线OFDM手势识别技术研究

Gesture recognition is the core of the newly-developed touchless Human-Computer interaction. The OFDM-based wireless naked hand gesture recognition can distinguish different gesture by detecting the time-variant doppler in OFDM due to the motion of hand. This method has the merit of ease of use, long sensing distance, being insensitive to temperature and illumination, and can even work in non-line-of-sight scenarios. However, to detect the micro-doppler caused by gesture in wireless channel, we have to face two major challenges. One is the heavy computation load required for processing the huge data used in micro-doppler detection, which makes realtime gesture recognition very difficult. The other problem is that gesture detection becomes unreliable due to interference and fading inherent in wireless channel. We reformulate the OFDM-based gesture recognition problem into two sub-problems: gesture design and fast doppler detection. For the former problem, we construct two fundamental gestures and map it into the notation of 0 and 1 bit as used in communication system. In this way, we can use two fundamental gestures to form new gestures by error-correction coding and thus reduce the probability of erroneous detection. For the latter, we use the idea of sparse fast fourier transform (SFFT) to deal with the high computation overhead as well as to detect gestures in realtime. If the above novel ideas work well, we will be able to provide a novel realtime and nearly error-free gesture recognition method, which may meet the rigorous requirement on detection time and reliabilty of those application, such as the control of bomb disposal robot and military command.

手势识别是新一代非接触式人机交互中的核心技术。基于OFDM信号的无线裸手手势识别技术通过检测OFDM信号中由人手运动引起的时变多普勒频移来辨别手势，具有使用便捷、作用距离远、受温度和亮度影响小、可应用于非视距环境等诸多优势。然而，在无线信道中探测由手势引起的微多普勒频移面临两大挑战：一是探测微多普勒需处理的数据量庞大，实时处理困难；二是无线信道中的干扰、衰落效应使可靠手势检测难以保证。本项目将基于OFDM的手势识别问题划分为手势设计和快速多普勒检测两个子问题。对前者，我们设计两个基本手势单元并将其映射为通信系统中0、1比特，然后利用基本手势单元对手势进行纠错编码以降低手势的误检测率；对后者，我们使用稀疏傅里叶变换降低多普勒检测的复杂度，提高手势检测的实时性。这一思路有望提供一种快速无损的手势检测方法，满足军事指挥、拆弹机器人控制等实时性和可靠性要求高的应用。

手势识别是新一代非接触式人机交互中的核心技术。基于OFDM信号的无线裸手手势识别技术通过检测OFDM信号中由人手运动引起的时变多普勒频移来辨别手势，具有使用便捷、作用距离远、受温度和亮度影响小、可应用于非视距环境等诸多优势。然而，在无线信道中探测由手势引起的微多普勒频移面临两大挑战：一是探测微多普勒需处理的数据量庞大，实时处理困难；二是无线信道中的干扰、衰落效应使可靠手势检测难以保证。针对上述问题以及进一步提高手势跟踪精度实现任意手势识别的需求，本项目主要研究了OFDM系统中快速Doppler 频移检测算法、高效的手势纠错编码设计、NLOS信道下OFDM整数频偏估计算法、快速精确的多频相位解缠方法和高精度的快速到达角估计（DOA）算法。项目将OFDM信号中微多普勒Doppler检测问题转化为谱估计问题，并使用稀疏傅里叶变换(SFFT)算法来实现谱估计，解决了稀疏傅里叶用于OFDM信号时的稳定性缺陷问题；项目首次将纠错编码思想用于手势识别，通过构造基本手势实现手势的纠错编码; 项目提出了一种在定时偏移和严重NLOS信道下，利用两块互素的导频完成整数频率估计(IFO)方法，算法在定时误差容忍度、NLOS信道和快变信道等环境下优于现有各种算法；项目提出了用于测距系统的具有闭合解的多频相位解缠方法``concerto''法，新方法可靠性高、复杂度极低、测距精度渐进于克拉美罗界，适用于多频OFDM测距、光学多波长光干涉测距及逆合成孔径雷达INSAR等多种测距系统。项目提出了一种高精度的快速到达角估计（DOA）P-to-P算法，该方法的信噪比门限低，复杂度很低，且适用于任意非线性阵列。项目提供一种快速无损的手势检测方法，满足军事指挥、拆弹机器人控制等实时性和可靠性要求高的应用。并且，项目研究的多频测距及DOA辅助测距技术有望实现高精度的手势跟踪，最终实现空中书写的目标。