Scholarlink

<p>Human gait analysis, a field focused on studying individuals' walking patterns, holds significant importance across various sectors due to its multifaceted applications and valuable insights into human movement dynamics. Whether in biometrics, healthcare, or surveillance, gait analysis provides critical information for identification, diagnosis, and monitoring purposes. However, traditional gait recognition systems encounter challenges such as variations in lighting conditions, occlusions, and changes in viewpoint, which can impair their accuracy and reliability. To address these challenges effectively, our study proposes an innovative method: the Improved Dense Capsule Network (IDCN). This groundbreaking approach represents a significant advancement in the field of gait analysis, combining deep learning techniques with the intricacies of human movement analysis. By aiming to enhance the precision and robustness of gait analysis systems, IDCN offers promising potential for overcoming limitations posed by variability in lighting conditions, occlusions, and changes in viewpoint. The process begins with collecting silhouette images from dataset and converting them into skeleton images using Media Pipe Pose (MPP), an open source platform provided by Google. Additionally, Convolutional neural networks (CNNs) are then employed to extract primary features from both silhouette images and skeleton images types. An Improved Dense CapsNet (IDCN) model is introduced, integrating high-level features from silhouette images and human poses for gait image recognition. The IDCN model customizes CapsNet by utilizing densely connected convolutions and incorporates an additional reconstruction loss to enhance capsule steering and encode instantiation parameters. The proposed scheme is implemented in Python, and its performance is evaluated in terms of Accuracy, Specificity, Precision, Recall, F-Score, FNR, FPR, computational time, and recognition time. The integration of silhouette and skeleton images enhances the robustness and accuracy of gait analysis systems, offering promising applications in biometrics, healthcare, and surveillance.</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

Deep Learning Empowered Human Gait Recognition with Improved Dense Capsule Networks

  • 期刊论文

  • 发表日期：

    2025年6月28日

查看全文
  • Arun Kumar Jhapate,
  • Hemang Shrivastava

摘要

Human gait analysis, a field focused on studying individuals' walking patterns, holds significant importance across various sectors due to its multifaceted applications and valuable insights into human movement dynamics. Whether in biometrics, healthcare, or surveillance, gait analysis provides critical information for identification, diagnosis, and monitoring purposes. However, traditional gait recognition systems encounter challenges such as variations in lighting conditions, occlusions, and changes in viewpoint, which can impair their accuracy and reliability. To address these challenges effectively, our study proposes an innovative method: the Improved Dense Capsule Network (IDCN). This groundbreaking approach represents a significant advancement in the field of gait analysis, combining deep learning techniques with the intricacies of human movement analysis. By aiming to enhance the precision and robustness of gait analysis systems, IDCN offers promising potential for overcoming limitations posed by variability in lighting conditions, occlusions, and changes in viewpoint. The process begins with collecting silhouette images from dataset and converting them into skeleton images using Media Pipe Pose (MPP), an open source platform provided by Google. Additionally, Convolutional neural networks (CNNs) are then employed to extract primary features from both silhouette images and skeleton images types. An Improved Dense CapsNet (IDCN) model is introduced, integrating high-level features from silhouette images and human poses for gait image recognition. The IDCN model customizes CapsNet by utilizing densely connected convolutions and incorporates an additional reconstruction loss to enhance capsule steering and encode instantiation parameters. The proposed scheme is implemented in Python, and its performance is evaluated in terms of Accuracy, Specificity, Precision, Recall, F-Score, FNR, FPR, computational time, and recognition time. The integration of silhouette and skeleton images enhances the robustness and accuracy of gait analysis systems, offering promising applications in biometrics, healthcare, and surveillance.