时空细粒度的跨模态无线感知方法研究

近些年来，研究者探索将广泛存在的WiFi通讯设备作为人体感知传感器，进行诸如动作识别、身份认证、室内定位等研究。现有基于WiFi信号的行为感知的研究存在着时间和空间的感知粒度不够精细的问题，无法完全满足细粒度感知应用需求。本项目从时间维度的持续动作识别、空间维度的三维人体姿态估计、空间维度的跨场景的人体高分辨率成像等三个研究点入手，利用图像等跨模态数据做标注，设计契合感知任务的信息编码方法和优化目标函数，构建符合无线数据特点的深度学习模块和算法，实现时空细粒度的人体感知研究，以推动WiFi感知技术朝更细粒度层次发展，为WiFi细粒度感知应用需求提供智能解决方案。

In recent years, researchers are exploiting the use of ubiquitous WiFi communication devices as human sensing sensors for activity recognition, identity authentication, indoor localization, etc. Existing WiFi-based works on behavior perception are coarse-grained in the aspect of time and space, not supporting usage scenarios that require fine-grained sensing results. To conduct temporal-spatially fine-grained sensing, in this project, we will utilize cross-modality information, design information coding methods suitable for perception tasks and optimize objective functions, and implement deep learning modules and algorithms that meet the characteristics of wireless data. The proposed approaches will be verified on three tasks, i.e., continuous activity recognition, 3D human pose reconstruction, and high-resolution imaging. The goal of this project is to promote the development of WiFi perception technology to a more fine-grained level, and provide intelligent solutions for the needs of fine perception applications.