1<!DOCTYPE HTML> 2 3<html> 4 <head> 5 <title>HandFi</title> 6 <meta charset="utf-8" /> 7 <meta name="viewport" content="width=device-width, initial-scale=1, user-scalable=no" /> 8 <link rel="stylesheet" href="assets/css/main.css" /> 9 <link rel="icon" href="images/logo.ico" type="image/x-icon"> 10 </head> 11 <body class="is-preload"> 12 13 <!-- Home Close--> 14 15 <div class="container"> 16 17 <h1 align="center"><strong> 3D Hand Skeleton with Commercial WiFi</strong></h1> 18 <br> 19 <h3 align="center"><strong>Sijie Ji<sup>1</sup>, Xuanye Zhang<sup>1</sup>, Yuanqing Zheng<sup>2</sup>, Mo Li<sup>3</sup> </strong></h3> 20 <p align="center"><sup>1</sup>Nanyang Technological University, 21 <sup>2</sup>The Hong Kong Polytechnic University, 22 <sup>3</sup>The Hong Kong University of Science and Technology</p> 23 <div style="text-align: center;"> 24 <a href="https://www4.comp.polyu.edu.hk/~csyqzheng/papers/HandFi_SenSys23.pdf" class="icon fa-file-pdf-o" style="font-size: 24px;"></a> 25 <a href="https://docs.google.com/presentation/d/1cmC4sboA6s1vlVGXur64SOQtPu47zg6d/edit?usp=drive_link&ouid=103151478846364717601&rtpof=true&sd=true" class="icon fa-file-powerpoint-o" style="font-size: 24px";></a> 26 <a href="https://www.youtube.com/watch?v=VNiWbpmgr5U" class="icon fa-file-video-o" style="font-size: 24px";></a> 27 <a href="https://github.com/SIJIEJI/HandFi" class="icon fa-file-code-o" style="font-size: 24px";></a> 28 </div> 29 <br> 30 <br> 31 <div style="text-align: center;"> 32 <iframe width="650" height="400" src="https://www.youtube.com/embed/VNiWbpmgr5U" frameborder="0" allowfullscreen></iframe> 33 </div> 34 <br> 35 <br> 36 <h2 align="center"><strong>Abstract</strong></h2> 37 <p align="center">This paper presents HandFi, which constructs hand skeletons with 38 practical WiFi devices. 39 Unlike previous WiFi hand sensing systems 40 that primarily employ predefined gestures for pattern matching, 41 by constructing the hand skeleton, HandFi can enable a variety 42 of downstream WiFi-based hand sensing applications in gaming, 43 healthcare, and smart homes. Deriving the skeleton from WiFi 44 signals is challenging, especially because the palm is a dominant 45 reflector compared with fingers. HandFi develops a novel multi-task 46 learning neural network with a series of customized loss functions 47 to capture the low-level hand information from WiFi signals. During 48 offline training, HandFi takes raw WiFi signals as input and uses 49 the leap motion to provide supervision. During online use, only 50 with commercial WiFi, HandFi is capable of producing 2D hand 51 masks as well as 3D hand poses. We demonstrate that HandFi can 52 serve as a foundation model to enable developers to build various 53 applications such as finger tracking and sign language recognition, 54 and outperform existing WiFi-based solutions.</p> 55 56 57 <!-- <div class="container is-max-desktop content"> --> 58 <div class="col-4 col-6-medium col-12-small"> 59 <section class="box style2"> 60 <h2 class="title">BibTeX</h2> 61 <code>@inproceedings{HandFi_SenSys23, 62 author = {Ji, Sijie and Zhang, Yuanye and Zheng, Yuanqing and Li, Mo}, 63 title = {Construct 3D Hand Skeleton with Commercial WiFi}, 64 year = {2023}, 65 publisher = {Association for Computing Machinery}, 66 booktitle = {Proceedings of the 21th ACM Conference on Embedded Networked Sensor Systems}, 67 series = {SenSys '23} 68 }</code> 69 70 </section> 71 </div> 72 73 74 75 </div> 76 77 78 <!-- Contact --> 79 <article id="contact" class="wrapper style4"> 80 <div class="container medium"> 81 <header> 82 83 <h3>Address</h3> 84 <p> 85 86 N4-B2a-01, School of Computer Science and Engineering (SCSE)</p> 87 <p> Nanyang Technological University, Singapore 639798</p> 88 </header> 89 <header> 90 <h3>Email</h3> 91 <p> [email protected] </p> 92 </header> 93 94 95 96 97 98 </div> 99 100 101 102 <!-- <div class = "col-12"> 103
103<script type="text/javascript" id="clustrmaps" src="//cdn.clustrmaps.com/map_v2.js?cl=ffffff&w=268&t=tt&d=32HsWfCBlA3pH7w0_vhA1pzvtzoDkcXu1bg9Hg3yTQo"></script>
103 104 </div> --> 105 <footer> 106 <ul id="copyright"> 107 © Sijie. Power by <a href = "https://html5up.net/" >HTML5 UP</a>. 108 </ul> 109 </footer> 110 </div> 111 </article> 112 113 <!-- Scripts --> 114
114<script src="assets/js/jquery.min.js"></script>
114 115
115<script src="assets/js/jquery.scrolly.min.js"></script>
115 116
116<script src="assets/js/browser.min.js"></script>
116 117
117<script src="assets/js/breakpoints.min.js"></script>
117 118
118<script src="assets/js/util.js"></script>
118 119
119<script src="assets/js/main.js"></script>
119 120 121 <!-- å¼å ¥jQueryåº --> 122<!--
122<script src="https://code.jquery.com/jquery-3.6.0.min.js"></script>
122 --> 123 124 125 126 </body> 127</html>
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.