基于Wi-Fi成像的连续动作识别研究论文.docx

基于Wi-Fi成像的连续动作识别研究 摘 要 摘 要 随着普适计算和物联网的快速发展，人体动作识别在越来越多的领域扮演了重要的角色，如智能家居、体感游戏、医疗看护等。这也给人体动作识别技术提出了新的要求，非绑定、易于部署、适用性强、不影响用户的连续动作识别系统成为技术发展的目标。针对上述目标，目前的方法均有所不足。例如，基于视频的方法虽然精确，但是侵犯用户隐私，会给用户带来不便；基于Wi-Fi的方法非绑定、易于部署，但是很难有效地分割连续动作。 针对上述问题，本文提出了一种新的解决思路，结合上述两个领域的优点，先进行Wi-Fi成像，再利用图像处理的方法进行连续动作分割和识别。对于Wi-Fi成像部分，本文根据身体不同部分相对于接收机处于不同的空间角度，创新性地将成像问题转化为角度评估问题，设计了一个基于圆阵和多载波的成像算法，通过Wi-Fi信号的子载波之间可测量的相位差，将子载波当成虚拟天线，来提高成像效果，并通过一种结合空间平滑算法和修正MUSIC算法的方法对相干信号解相干，进而得到空间中每个方位角和俯仰角的信号强度，即人体热图。对于图像处理部分，针对本文中成像热图和人体动作的特点，提出了一种基于包围盒面积变化的连续动作分割方法进行分割，然后使用PCA降维，最后使用SVM进行动作识别。 本文使用USRP硬件平台搭建了该系统。实验表明，本文提出的方法可以实现较好的成像效果，对5个不同动作的识别准确率达到92.6%，并且模型的鲁棒性较高，适用于多种场景，例如用户体型不同、测试距离不同、身体旋转角度不同等情况。 关 键 词：Wi-Fi；MUSIC；成像；动作识别 论文类型：应用研究 西安交通大学硕士学位论文 ABSTRACT PAGE 14 ABSTRACT With the rapid development of pervasive computing and internet of things, human activity recognition plays an important role in more and more fields, such as smart home, somatosensory games, medical care, etc. It also brings new requirements for human activity recognition technology. The continuous activity recognition system, which is device-free, deployable, universal and does not affect users, has become the goal of technological development. For the above objective, there are some shortcomings in the current methods. For example, although the video-based methods are accurate, they cause inconvenience to users and invades user’s privacy. The Wi-Fi-based methods are device-free and deployable, but it is difficult to effectively segment continuous actions. In order to solve the above problems, a new solution is proposed in this paper. Combining the advantages of the above two areas, Wi-Fi imaging is first performed, and then continuous activity segmentation and recognition are carried out by image processing method. For the Wi-Fi imaging part, according to different parts of the body are at different spatial angles relative to the receiver, this paper innovatively transforms the Wi-Fi imaging into the angle estimation, and designs an im