Multi-rate-Directed-T-GCN

Aggregate Together We See Through: 
WiFi-based Through-wall2D Human Pose Estimation via Multi-rate Directed T-GCN

• Architecture

• Dataset

• Visulization

• Result

Architecture

Dataset

Place	Type	Url	Place	Type	Url
P1	single-person 01	https://reurl.cc/oLE0bV	P1	single-person 02	https://reurl.cc/E75GNa
P1	single-person 03	https://reurl.cc/MvKRmm	P1	single-person 04	https://reurl.cc/ZOLWjM
P1	single-person 05	https://reurl.cc/z8nYW6	P1	single-person 06	https://reurl.cc/1xN1oV
P1	multi-people	https://reurl.cc/9EmVA8	P1	wall	https://reurl.cc/WdKDY5
P2	single-person 01	https://reurl.cc/5l9MyR	P2	single-person 02	https://reurl.cc/62nN3k
P2	single-person 03	https://reurl.cc/j7Klj2	P2	single-person 04	https://reurl.cc/b5K9V6
P2	single-person 05	https://reurl.cc/3DqxXL	P2	single-person 06	https://reurl.cc/exKDzm
P2	multi-people	https://reurl.cc/4RVQ7V

Visulization

Our demo for wifi based 2D human pose estimation

Proposed Model	Person-in-WiFi[2]

Demo

multi-people single-person

Gap between Camera-based(Openpose) and Labeld Ground Truth. (Ramdomly sample 100 examples)

Result

The PCK@20(Percentage of Correct Keypoint)of provided models are shown here:

Method	single-person	multi-people
WiSPPN[1]	69.82%	X
Person-in-WiFi[2]	77.06%	61.58%
MDT-GCN(ours)	82.26%	71.58%

Method	through-wall
WiSPPN[1]	58.86%
Person-in-WiFi[2]	73.67%
MDT-GCN(ours)	80.72%

Gap between Camera-based(Openpose) and Labeld Ground Truth.(Ramdomly sample 100 examples)

Camera-based(Openpose)
100%

[1] Fei Wang, Stanislav Panev, Ziyi Dai, Jinsong Han, and Dong Huang. 2019. Canwifi estimate person pose?arXiv preprint arXiv:1904.00277(2019).

[2] Fei Wang, Sanping Zhou, Stanislav Panev, Jinsong Han, and Dong Huang. 2019.Person-in-WiFi: Fine-grained person perception using WiFi. InProceedings of theIEEE International Conference on Computer Vision. 5452–5461.