Wi-Fi CSI AoA Pose Estimation Without Cameras or mmWave
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing human pose estimation technologies require additional hardware, such as cameras or expensive millimeter wave devices, and pose estimation methods using RF signals or millimeter wave radar, which can violate privacy or pose health risks, and are costly and difficult to deploy widely.
Innovation Solution
A static human pose estimation method using CSI signal angle of arrival estimation, utilizing a teacher-student network to process CSI data collected through widely deployed Wi-Fi devices, constructing a two-dimensional angle of arrival image, and reducing environmental interference to achieve high prediction accuracy and low cost.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If visual methods (cameras) are used for human pose estimation, then pose estimation capability is achieved, but privacy leakage risk increases and user resistance increases
Solution Approach 1:
The patent replaces visual sensing (cameras) with electromagnetic wave sensing (CSI signals from Wi-Fi devices). Instead of using optical fields to capture images, the system uses radio frequency electromagnetic waves to detect human poses through phase information changes in CSI data, thereby avoiding privacy leakage while maintaining pose estimation capability
Solution Approach 2:
The patent introduces CSI signals as an intermediary between the sensing system and human pose. Rather than directly capturing visual images, the system uses Wi-Fi CSI phase information as a mediator to indirectly detect human pose, which preserves privacy while enabling accurate pose estimation through environmental interaction analysis
2Measurement precision
If millimeter wave radar is used for human pose estimation, then 3D skeleton keypoints and activity trajectories can be estimated, but device cost increases and health safety concerns arise due to high signal transmission power
Solution Approach 1:
The patent replaces expensive millimeter wave radar devices with widely deployed, low-cost commercial Wi-Fi devices. Instead of using specialized high-power radar hardware, the system leverages existing Wi-Fi infrastructure that transmits at much lower power levels, reducing both device cost and potential health risks while maintaining pose estimation functionality
Solution Approach 2:
The patent makes the system universally deployable by using common Wi-Fi devices that already exist in most environments, rather than requiring specialized millimeter wave radar equipment. The Wi-Fi devices perform multiple functions including communication and pose estimation, eliminating the need for additional dedicated sensing hardware
3Measurement precision
If RF signal-based methods are used for human pose estimation, then human pose can be estimated by analyzing signal changes, but expensive custom devices and professional deployment are required
Solution Approach 1:
The patent enables the system to use existing Wi-Fi infrastructure that already provides communication services. The Wi-Fi devices perform pose estimation as a secondary function using their existing transmission and reception capabilities, eliminating the need for separate dedicated sensing devices and professional deployment teams
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method provides non-contact, high-accuracy human pose estimation with low cost by using CSI data from commercial Wi-Fi devices, achieving an average keypoint prediction rate of 85.5% across different environments and improving accuracy by 18% compared to existing methods.
Implementation Method 1
δ is the propagation delay caused by multipath propagation
Data Source
AI summary
A static human pose estimation method based on channel state information (CSI) signal angle of arrival (AoA) estimation is disclosed. A signal angle of arrival in a sensing area is estimated from CSI and an image is constructed, and a human pose is estimated from the image using a teacher-student network. The method comprises the following specific steps: CSI signals in a sensing area are collected at different heights using a receive antenna column with a moving track. Two-dimensional AoA image features are constructed: CSI information is converted into one-dimensional AoA data by using a MUSIC algorithm, and the one-dimensional AoA data of different heights are combined into a two-dimensional AoA image, and an environmental denoise algorithm is designed. A teacher-student network model is constructed, and a student network based on the two-dimensional AoA image is supervised, wherein the student network model can independently estimate a human pose from CSI.