Scholarlink

<p>The study describes a strong gait-based biometric recognition system applied to the AGH Krakow Gait Dataset that includes gait numeric features like step length, cadence, and joint paths. To overcome intra-subject variability and guarantee temporal consistency, Z-score normalization and gait cycle segmentation with Temporal Segment Networks (TSN) are used. Biomechanical modeling allows the extractionof dynamic gait parameters, and this feature increases feature richness. The best discriminative features are chosen by a Hybrid Entropy-Driven Whale Optimization Algorithm (HEWOA) that minimizes redundancy. To classify, a Spatiotemporal Graph-Boost Classifier (ST-GBC) is designed, which represents both spatial and temporal gait patterns by graph structures and gradient boosting. The biometric metrics employed to test the framework with the 2 datasets, such as the proposed HEWOA model, are used to show the high generalization capability. On OU-MVLP, it records the best accuracy of 98.87, and Normal Walking Variations also record competitive results with 97.54 and 96.89, respectively. Precision, recall, and F1-score values continue to be over 96 with a very low FNR and FPR (&lt; 1.2%). These findings validate the strength, dependability, and flexibility of the model in various gait recognition tasks. This combined strategy shows that real-time gait authentication can be realized in both surveillance and healthcare uses, based on numerical information as compared to image-based information.</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

An optimized machine learning framework for spatiotemporal gait analysis and biometric recognition

  • 期刊论文

  • 发表日期：

    2025年12月27日

查看全文
  • Monika Jhapate,
  • Hemang Shrivastava

摘要

The study describes a strong gait-based biometric recognition system applied to the AGH Krakow Gait Dataset that includes gait numeric features like step length, cadence, and joint paths. To overcome intra-subject variability and guarantee temporal consistency, Z-score normalization and gait cycle segmentation with Temporal Segment Networks (TSN) are used. Biomechanical modeling allows the extractionof dynamic gait parameters, and this feature increases feature richness. The best discriminative features are chosen by a Hybrid Entropy-Driven Whale Optimization Algorithm (HEWOA) that minimizes redundancy. To classify, a Spatiotemporal Graph-Boost Classifier (ST-GBC) is designed, which represents both spatial and temporal gait patterns by graph structures and gradient boosting. The biometric metrics employed to test the framework with the 2 datasets, such as the proposed HEWOA model, are used to show the high generalization capability. On OU-MVLP, it records the best accuracy of 98.87, and Normal Walking Variations also record competitive results with 97.54 and 96.89, respectively. Precision, recall, and F1-score values continue to be over 96 with a very low FNR and FPR (< 1.2%). These findings validate the strength, dependability, and flexibility of the model in various gait recognition tasks. This combined strategy shows that real-time gait authentication can be realized in both surveillance and healthcare uses, based on numerical information as compared to image-based information.