Scholarlink

<p>Automated surveillance systems use deep learning techniques for person re-identification across multi-camera networks, overcoming limitations in single-domain environments and pseudo-label noise using DCNNs. This study proposes a novel model that combines image quality enhancement, feature extraction and fusion, dimensionality reduction, and DCNN-based identification. The model employs data preprocessing using the Geometric Wiener Filter (GWF) and Histogram Equalization (HE), feature extraction via global and local features, multi-feature fusion with the Multi-Feature Fusion Model (MFFM), and dimension reduction using Principal Component Analysis (PCA). The DCNN is then utilized for person identification. The model’s effectiveness is validated against existing methods, such as DNN, CNN, and Bi-LSTM, using various metrics on datasets like Market1501 and Duke-MTMC-Re-ID, demonstrating superior performance with an accuracy rate of 99%. The model is implemented in MATLAB for thorough testing and simulation of real-world scenarios, showcasing its robustness and efficiency in cross-domain applications. The proposed methodology collectively enhances the model’s capability to perform effective person re-identification in diverse and challenging environments, addressing common issues faced in cross-domain scenarios and preserving subjects’ identities while improving generalization and privacy in visual data recognition.</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

Cross-domain person re-identification with deep convolutional neural network

  • 期刊论文

  • 发表日期：

    2026年2月25日

查看全文
  • Malliga Padmanaban,
  • Manjula Dhanabalachandran

摘要

Automated surveillance systems use deep learning techniques for person re-identification across multi-camera networks, overcoming limitations in single-domain environments and pseudo-label noise using DCNNs. This study proposes a novel model that combines image quality enhancement, feature extraction and fusion, dimensionality reduction, and DCNN-based identification. The model employs data preprocessing using the Geometric Wiener Filter (GWF) and Histogram Equalization (HE), feature extraction via global and local features, multi-feature fusion with the Multi-Feature Fusion Model (MFFM), and dimension reduction using Principal Component Analysis (PCA). The DCNN is then utilized for person identification. The model’s effectiveness is validated against existing methods, such as DNN, CNN, and Bi-LSTM, using various metrics on datasets like Market1501 and Duke-MTMC-Re-ID, demonstrating superior performance with an accuracy rate of 99%. The model is implemented in MATLAB for thorough testing and simulation of real-world scenarios, showcasing its robustness and efficiency in cross-domain applications. The proposed methodology collectively enhances the model’s capability to perform effective person re-identification in diverse and challenging environments, addressing common issues faced in cross-domain scenarios and preserving subjects’ identities while improving generalization and privacy in visual data recognition.