Scholarlink

<p>Person identity recognition from obscured video traces is a necessary precondition for recognizing individuals from surveillance camera outputs. High-level tasks such as person/object recognition, analysis, tracking, tracing, and monitoring depend on accurate identity recovery. Identifying the same person across multiple frames of obscured video sequences is referred to as person re-identification (ReID). These identity traces may originate from a single camera or a network of spatially distributed non-overlapping cameras. Motivated by recent advancements in Convolution Neural Networks (CNN) and Radial Basis Function Neural Networks (RBFNN), we propose a novel hybrid architecture called Hybrid Multi-Layer Perceptron Radial Basis Function Neural Network (HMLP-RBFNN), constructed by modifying standard CNN and RBFNN architectures. The performance of the proposed model is evaluated on publicly available datasets. Experimental analysis demonstrates the superiority of the proposed approach. The standalone CCNN model achieves re-identification Rank-1 accuracy of 91.10%, 92.50%, 91.28%, and 94.83% on the MARS, RPIfield, LaST, and DeepChange datasets, respectively. The hybrid HMLP-RBFNN pipeline achieves improved Rank-1 accuracy of 98.20%, 97.50%, 95.78%, and 97.10% on the same datasets, significantly outperforming the standalone model. All results are obtained using identity-disjoint training/testing splits and evaluated using standard CMC and mAP metrics.</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

A combined approach - recognizing subjects after obstruction using hybrid multilayer perceptron’s kernels

  • 期刊论文

  • 发表日期：

    2026年7月2日

查看全文
  • N. A. Deepak

摘要

Person identity recognition from obscured video traces is a necessary precondition for recognizing individuals from surveillance camera outputs. High-level tasks such as person/object recognition, analysis, tracking, tracing, and monitoring depend on accurate identity recovery. Identifying the same person across multiple frames of obscured video sequences is referred to as person re-identification (ReID). These identity traces may originate from a single camera or a network of spatially distributed non-overlapping cameras. Motivated by recent advancements in Convolution Neural Networks (CNN) and Radial Basis Function Neural Networks (RBFNN), we propose a novel hybrid architecture called Hybrid Multi-Layer Perceptron Radial Basis Function Neural Network (HMLP-RBFNN), constructed by modifying standard CNN and RBFNN architectures. The performance of the proposed model is evaluated on publicly available datasets. Experimental analysis demonstrates the superiority of the proposed approach. The standalone CCNN model achieves re-identification Rank-1 accuracy of 91.10%, 92.50%, 91.28%, and 94.83% on the MARS, RPIfield, LaST, and DeepChange datasets, respectively. The hybrid HMLP-RBFNN pipeline achieves improved Rank-1 accuracy of 98.20%, 97.50%, 95.78%, and 97.10% on the same datasets, significantly outperforming the standalone model. All results are obtained using identity-disjoint training/testing splits and evaluated using standard CMC and mAP metrics.