US 12,481,055 B2
Static human pose estimation method based on CSI signal angle of arrival estimation
Fu Xiao, Jiangsu (CN); Mingming Xu, Jiangsu (CN); Zhengxin Guo, Jiangsu (CN); Hai Hu, Jiangsu (CN); Linqing Gui, Jiangsu (CN); Biyun Sheng, Jiangsu (CN); Jian Zhou, Jiangsu (CN); and Hui Cai, Jiangsu (CN)
Assigned to Nanjing University of Posts and Telecommunications, Jiangsu (CN)
Appl. No. 18/271,236
Filed by Nanjing University of Posts and Telecommunications, Jiangsu (CN)
PCT Filed Oct. 13, 2022, PCT No. PCT/CN2022/125127
§ 371(c)(1), (2) Date Jul. 7, 2023,
PCT Pub. No. WO2023/213051, PCT Pub. Date Nov. 9, 2023.
Claims priority of application No. 202210484261.X (CN), filed on May 6, 2022.
Prior Publication US 2025/0155567 A1, May 15, 2025
Int. Cl. G01S 13/89 (2006.01); G01S 5/02 (2010.01); G01S 13/87 (2006.01); G06T 7/70 (2017.01)
CPC G01S 13/89 (2013.01) [G01S 5/0252 (2013.01); G01S 13/87 (2013.01); G06T 7/70 (2017.01); G06T 2207/30196 (2013.01)] 6 Claims
OG exemplary drawing
 
1. A static human pose estimation method based on CSI signal angle of arrival estimation, the human pose estimation method comprising the following steps:
Step 1: placing a receive antenna column with a moving track in a sensing area, using a fixed transmit antenna to send Wi-Fi data packets to the receive antenna mounted on the moving track, and moving the receive antenna to multiple specified heights to collect multiple CSI data at the respective heights and synchronously collect image data;
Step 2: extracting phase information in the CSI data, constructing eight one-dimensional angle of arrival (AoA) image, and combining the eight one-dimensional AoA images of different heights into a two-dimensional AoA image;
Step 3: reducing environmental interference factors in the two-dimensional AoA image by using an environmental denoise algorithm;
Step 4: inputting the image data into a teacher network to obtain supervised data of coordinates of human skeleton keypoints, and inputting the supervised data and the two-dimensional AoA image after denoise into a student network for training;
Step 5: upon predicting a human pose, placing a receive antenna column with a moving track in a sensing area, using a fixed transmit antenna to send Wi-Fi data packets to the receive antenna mounted on the moving track, moving the receive antenna to multiple specified heights, collecting one piece of CSI data at each height, and after extracting features through steps 2 and 3 from the CSI data collected at different heights and interpolating the features, inputting them into the student network model trained in step 4, to output predicted coordinates of the human skeleton keypoints of an object in the sensing area.