Abstract
The study of human gait recognition has been becoming an active research field. In this paper, we propose to adopt the attention-based Recurrent Neural Network (RNN) encoder-decoder framework to implement a cycle-independent human gait and walking direction recognition system in Wi-Fi networks. For capturing more human walking dynamics, two receivers together with one transmitter are deployed in different spatial layouts. In the proposed system, the Channel State Information (CSI) measurements from different receivers are first gathered together and refined to form an integrated walking profile. Then, the RNN encoder reads and encodes the walking profile into primary feature vectors. Given a specific recognition task, the decoder computes a corresponding attention vector which is a weighted sum of the primary features assigned with different attentions, and is finally used to predict the target. The attention scheme motivates our system to learn to adaptively align with different critical clips of CSI data sequence for human walking gait and direction recognitions. We implement our system on commodity Wi-Fi devices in indoor environment, and the experimental results demonstrate that our system can achieve average F1 scores of 89.69% for gait recognition from a group of 8 subjects and 95.06% for direction recognition from 8 walking directions, in addition, the average accuracies of these two recognition tasks both exceed 97%.
Introduction
Identifying individuals by their unique walking patterns—gait recognition—is a powerful tool for smart environments and healthcare monitoring. Traditional methods often rely on cameras, which raise significant privacy concerns and require clear lines of sight, or wearables, which can be intrusive for daily use. Wi-Fi-based sensing, specifically using Channel State Information (CSI), offers a promising "device-free" alternative. However, existing Wi-Fi systems often struggle with signal noise and the difficulty of accurately segmenting individual steps (gait cycles) from messy data.
This paper introduces a system that is both cycle-independent and multi-task, capable of identifying a person and their walking direction simultaneously without needing to perfectly isolate individual steps.
System Architecture and Physical Setup
The system leverages the way human movement distorts ambient Wi-Fi signals. To capture the full complexity of human kinetics, the researchers deployed a transmitter (Tx) and two receivers (Rx) at different heights (0.5m and 1.0m). This spatial diversity allows the system to capture distinct dynamics: the lower receiver focuses on leg movements, while the higher receiver picks up torso and arm swings.
Raw CSI data is processed through a pipeline involving Butterworth bandpass filters to isolate human-movement frequencies (5–90 Hz) and Principal Component Analysis (PCA) to extract the most salient signal variations, forming an integrated walking profile.
The Attention-based RNN Framework
The core innovation of this system is the attention-based RNN encoder-decoder framework. Unlike standard RNNs that compress an entire sequence into a single fixed vector, the attention mechanism allows the decoder to "look back" and assign different weights to specific parts of the walking sequence.
This mechanism enables the system to focus on the most relevant "clips" of movement for a given task. For instance, the model might prioritize the peak of a leg swing for identity recognition while focusing on broader signal trends to determine walking direction. This approach makes the system robust to pauses or partial steps, as it does not rely on rigid gait cycle detection.
Experimental Evaluation and Results
The system was tested in a laboratory environment with 11 volunteers across 12 walking paths. The results demonstrate high precision across both primary tasks:
- Gait Recognition: Achieved an average accuracy exceeding 97% and an F1-score of 89.69% for a group of 8 subjects.
- Direction Recognition: Achieved an average accuracy exceeding 97% and an F1-score of 95.06% across 8 different walking directions.
By deploying receivers at different heights, the system gains a "stereo" view of human kinetics, significantly improving its ability to distinguish between individuals compared to single-receiver setups.
Conclusion and Future Work
This research proves that attention-based deep learning can overcome the traditional hurdles of Wi-Fi sensing, such as noise and the need for manual feature engineering. While highly effective for single-person scenarios, future work will focus on untangling "signal collisions" that occur when multiple people walk simultaneously, as well as testing the system's robustness against environmental changes like furniture layout and different types of footwear.
Summary updated