%0 Journal Article
%T TFIE-Gait：一种基于时频信息增强的步态识别模型
TFIE-Gait: A Gait Recognition Model Based on Temporal-Frequency Information Enhancement
%A 蔡俊
%J Modeling and Simulation
%P 858-870
%@ 2324-870X
%D 2025
%I Hans Publishing
%R 10.12677/mos.2025.144336
%X 在步态识别任务中，空间和时序信息对区分不同步态模式至关重要。然而，现有方法在开放环境数据集(如Grew)中主要依赖空间信息，未充分利用时序信息，且开放环境数据集中的噪声(如遮挡和运动暂停)会破坏步态序列时序信息，干扰时序特征提取，降低模型性能。为此，本文提出TFIE-Gait模型，引入时频信息增强模块(TFIE)和去噪采样模块(DAS)。TFIE模块结合时域和频域信息，通过多尺度卷积和自注意力机制提取关节时序特征及关节间依赖关系，并利用傅里叶变换在频域提取判别性特征。DAS模块利用频域去噪前后的序列数据差异联合分析，识别和去除异常数据帧，并利用交叉相关算法拼接子序列，恢复步态序列的周期性时序信息。实验表明，TFIE-Gait在开放环境数据集上显著优于基线模型。
In gait recognition tasks, spatial and temporal information are crucial for distinguishing different gait patterns. However, existing methods primarily rely on spatial information in open-environment datasets (e.g., GREW) and fail to fully utilize temporal information. Moreover, noise in open-environment datasets (e.g., occlusions and motion pauses) can disrupt the temporal information of gait sequences, interfere with temporal feature extraction, and degrade model performance. To address these issues, this paper proposes the TFIE-Gait model, which introduces a Time-Frequency Information Enhancement (TFIE) module and a Denoising and Sampling (DAS) module. The TFIE module integrates time-domain and frequency-domain information, leveraging multi-scale convolution and self-attention mechanisms to extract joint temporal features and inter-joint dependencies, while utilizing Fourier transform to extract discriminative features in the frequency domain. The DAS module jointly analyzes the differences between sequences before and after frequency-domain denoising to identify and remove abnormal data frames, and employs a cross-correlation algorithm to stitch subsequences, thereby restoring the periodic temporal information of gait sequences. Experimental results demonstrate that TFIE-Gait significantly outperforms baseline models on open-environment datasets.
%K 步态识别，
%K 时频分析，
%K 时序建模，
%K 自注意力机制
Gait Recognition
%K Time-Frequency Analysis
%K Temporal Modeling
%K Self-Attention Mechanism
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=112826