Publication

Abstract : Gait-based human identification has gained significant attention as a reliable biometric approach due to its non-intrusive nature and ability to identify individuals from a distance. Recent advancements in computer vision and deep learning have improved gait recognition performance; however, challenges related to noisy data, inaccurate segmentation, and classification complexity still affect system reliability. To address these issues, this paper presents an improved score-level fused hybrid classifier for gait-based human identification (ISFHC-GHI). Initially, the input gait frames are preprocessed using an adaptive LOG-based Wiener filtering (AdLOG-WF) approach to suppress noise while preserving important image details. The preprocessed frames are then segmented using a Modified U-Net (mU-Net) model designed to retain fine structural information and improve boundary segmentation. Subsequently, multiple discriminative features, including Hierarchy of Skeleton (HOS), Multitexton, and color-based features, are extracted to capture structural, textural, and appearance-related gait characteristics. The extracted features are classified using a hybrid framework based on LinkNet and SqueezeNet classifiers. Finally, an improved score-level fusion strategy combines the classifier outputs to enhance identification performance and classification stability. Experimental analysis conducted on benchmark gait datasets demonstrates that the proposed ISFHC-GHI framework achieves improved performance compared to existing methods, attaining an accuracy of 0.915, precision of 0.870, and NPV of 0.937.