Scholarlink

<p>Gait recognition has gained prominence as a contactless biometric technique that identifies individuals by analyzing their distinctive walking styles. Unlike conventional biometric systems, this approach can function from a distance and does not require active cooperation from the subject, making it especially suitable for applications in surveillance, security, and healthcare. Despite its advantages, the technique faces several challenges, including changes in walking scenarios, camera viewpoints, and variations in attire, all of which can hinder accuracy. In this study, an optimized gait recognition framework is introduced, utilizing the CASIA-B dataset along with feature extraction through two pre-trained convolutional neural networks: AlexNet and MobileNet. The extracted features are classified using a Multi-Layer Perceptron (MLP). The results reveal that AlexNet outperforms MobileNet, achieving a peak accuracy of 93.51% with a minimal loss of 2.15% at an 18° angle of view. Further analysis using performance indicators such as Precision, Recall, F1-score, and Loss underscores the reliability and effectiveness of the proposed model. The findings demonstrate the advantages of combining deep learning models for improved recognition performance under diverse conditions. Future efforts will aim at implementing real-time recognition capabilities and scaling the approach to broader datasets for enhanced generalization.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced Gait Recognition Under Covariate Conditions Using Pre-Trained CNNs and Multi-Layer Perceptron

  • 期刊论文

  • 发表日期：

    2026年3月12日

查看全文
  • Mohd Shukri Ab Yajid,
  • Vivek Vullikanti,
  • Deeksha Verma,
  • Ahmed Alkhayyat,
  • Lakshay Bareja,
  • Mandeep Kaur Chohan,
  • Manoranjan Parhi,
  • Devendra Singh

摘要

Gait recognition has gained prominence as a contactless biometric technique that identifies individuals by analyzing their distinctive walking styles. Unlike conventional biometric systems, this approach can function from a distance and does not require active cooperation from the subject, making it especially suitable for applications in surveillance, security, and healthcare. Despite its advantages, the technique faces several challenges, including changes in walking scenarios, camera viewpoints, and variations in attire, all of which can hinder accuracy. In this study, an optimized gait recognition framework is introduced, utilizing the CASIA-B dataset along with feature extraction through two pre-trained convolutional neural networks: AlexNet and MobileNet. The extracted features are classified using a Multi-Layer Perceptron (MLP). The results reveal that AlexNet outperforms MobileNet, achieving a peak accuracy of 93.51% with a minimal loss of 2.15% at an 18° angle of view. Further analysis using performance indicators such as Precision, Recall, F1-score, and Loss underscores the reliability and effectiveness of the proposed model. The findings demonstrate the advantages of combining deep learning models for improved recognition performance under diverse conditions. Future efforts will aim at implementing real-time recognition capabilities and scaling the approach to broader datasets for enhanced generalization.