Poster: MilliPose: Facilitating Full Body Silhouette Imaging from Millimeter-Wave Device

Poster: MilliPose: Facilitating Full Body Silhouette Imaging from Millimeter-Wave Device
复制标题

海报:MilliPose:促进毫米波设备的全身轮廓成像

DOI:
--
复制
发表时间:
2021
期刊:
影响因子:
--
通讯作者:
Sanjib Sur
Sanjib Sur
中科院分区:
--
文献类型:
--
作者:
Aakriti Adhikari;Sanjib Sur

文献摘要

参考文献

被引文献

相似文献

本文提出了MilliPose系统,该系统有助于利用毫米波(mmWave)设备实现人体全身轮廓成像和三维姿态估计。与现有的基于视觉的运动捕捉系统不同,MilliPose不会侵犯隐私,并且能够在有遮挡、能见度差和低光照条件下工作。MilliPose利用基于条件生成对抗网络和循环神经网络的机器学习模型,来应对现有毫米波成像系统分辨率低、存在镜面反射以及反射率多变等挑战。我们的初步结果表明,MilliPose在准确预测自然人体运动下的身体关节位置方面行之有效。
This paper proposes MilliPose , a system that facilitates full human body silhouette imaging and 3D pose estimation from millimeter-wave (mmWave) devices. Unlike existing vision-based motion capture systems, MilliPose is not privacy-invasive and is capable of working under obstructions, poor visibility, and low light conditions. MilliPose leverages machine-learning models based on conditional Generative Adversarial Networks and Recurrent Neural Network to solve the challenges of poor resolution, specularity, and variable reflectivity with existing mmWave imaging systems. Our prelimi-nary results show the efficacy of MilliPose in accurately predicting body joint locations under natural human movement.
ZigZagCam:突破手持式毫米波成像的极限
DOI: --
发表时间: 2021
期刊: Proceedings of the 22nd International Workshop on Mobile Computing Systems and Applications
影响因子: --
作者:
Regmi, Hem;Saadat, Moh Sabbir;Sur, Sanjib;Nelakuditi, Srihari
通讯作者: Nelakuditi, Srihari