基于毫米波设备的远距离多目标追踪算法研究  

Research on Long-Range Multi-Target Tracking Algorithm with Millimeter-Wave Device

作　　者：张桂栋 杨铮[1] 张驿 迟国轩 马强 苗欣 ZHANG Gui-Dong;YANG Zheng;ZHANG Yi;CHI Guo-Xuan;MA Qiang;MIAO Xin(School of Software,Tsinghua University,Beijing 100084)

机构地区：[1]清华大学软件学院,北京100084

出　　处：《计算机学报》2023年第7期1366-1382,共17页Chinese Journal of Computers

基　　金：国家自然科学基金(61832010,62072272)资助。

摘　　要：各类定位追踪方法的发展,促进了各类应用的兴起,这其中包括智能家居、虚拟现实以及安全监控.无线定位追踪算法也是物联网技术、乃至计算机领域研究的重点方向之一.相比于视觉、激光等方法,基于无线信号的定位追踪方法,由于受环境影响小,并且能够有效地保护人员的隐私信息,因而逐渐受到研究者们的关注.部分方法不需要目标携带特定的硬件设备,就能获取移动目标在不同时刻的位置.然而,大部分的方法不能满足部分工业应用的需求,这些工业应用要求在超过10m的距离下能够达到厘米级别的精度.相比于近距离的情况,远距离的定位追踪面临着两个亟待解决的挑战:一方面,当目标距离设备较远时,周围的静态物体以及目标的二次反射等干扰将会增强,这对定位系统的性能造成了严重的影响;另一方面,多个目标可能同时出现在监控区域中,它们的反射混叠在一起,使得多目标的定位更加困难.在本文中,我们提出了mi-Track,这是第一个基于毫米波设备实现设备无依赖的多目标远距离追踪并达到厘米级精度的方法.mi-Track工作在商用毫米波设备上,并且通过提取距离速度谱进行追踪.为了减少静态物体、二次反射等干扰的影响,mi-Track提出了移动目标敏感的粒子滤波算法,来实现单目标远距离高精度的连续追踪.考虑到移动过程中距离和速度的连续性,在追踪过程中我们同时利用了毫米波设备的量测值和上一时刻的预测值来调整粒子,并通过粒子的初始化策略使得系统更关注于移动目标,从而进一步减少了环境干扰的影响.在此基础上,我们依次检测多个目标并通过轨迹融合匹配算法,对属于单个目标的多段轨迹进行合并,实现了多目标的计数和追踪.我们将mi-Track部署到了商用毫米波设备上,并在三个场景中进行实验.实验结果表明,mi-Track的平均定位误差为2.4cm,相比Accurate location awareness stimulates a wide range of applications,e.g.,smart homes,virtual reality,and security surveillance.Therefore,wireless locating and tracking algorithm is also one of the key research directions of Internet of Things,and even computing.Compared to vision-based and laser-based methods,wireless localization and tracking methods have been emerging due to less environmental impact and non-invasiveness.Device-free tracking does not need any on-body equipment and exploits the reflected signal from the target for capturing the location of moving targets at different moments.Despite that existing localization systems achieve high and reliable accuracy in near-field scenarios,most are not capable of long-range target acquisition.Specifically,they do not meet the requirement of most industrial applications with a working range of more than ten meters and centimeter-level localization accuracy,such as traveling crane control,mechanical arm system,and train communication network design.Compared to close-range localization,long-range localization faces two main unresolved challenges which need to be addressed in the literature.On one hand,the sensing environment will be more complex within a longer range.As the target is away from the sensor,reflections from static objects at similar distances and secondary reflections from the target become comparable to the target’s direct reflection,which might deteriorate localization accuracy sharply.On the other hand,multiple targets can be in the monitoring area simultaneously.Their reflections will superimpose at the sensor,making the localization of multiple targets harder.In this paper,we propose mi-Track,the first method that passively tracks multiple targets in a long range with centimeter-level accuracy and outperforms the state of the arts.mi-Track works on commercial mmWave devices that transmit and receive frequency-modulated continuous wave signals,and utilizes the extracted normalized range-velocity spectrum for tracking.In order to relieve the in

关 键 词：无线感知 毫米波 目标追踪 调频连续波 远距离 多目标 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]