Scholarlink

<p>Person Re-identification (Re-ID) is employed for the identification of persons in several cameras. However, the process of detecting the person is not an easy task, because of the lack of database, and distance. Therefore, a new model known as Hybrid DeepID FaceNet (HyDeepIDF-Net) is employed for person Re-ID. At first, input video obtained from the dataset is allowed for frame extraction for the extraction of key frames from the input video. Next, face detection is done by Viola-Jone’s algorithm from the extracted keyframes. Finally, the residual image extraction from the detected face image is performed by employing Deep Residual Network (DRN). Similarly, the acquired face image for query is separately allowed for residual image extraction using DRN. Then, the resultant extracted residual images from query and input images are allowed for semantic matching using the proposed HyDeepIDF-Net. The HyDeepIDF-Net is created by accumulating DeeepID-Net and FaceNet. Here, DeepId Net and FaceNet are employed as they can find the similarities and differences between faces. At last, the tracked persons are annotated on the frames. Further, the assessment of HyDeepIDF-Net in terms of recognition percentage, accuracy, precision, F1-score, and recall show that maximum values of 92.12%, 93.91%, 91.41%, 92.32%, and 93.24%, and minimum latency of 0.54s are recorded.</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

HyDeepIDF-Net: Hybrid Deep ID FaceNet for Person Re-Identification

  • 期刊论文

  • 发表日期：

    2025年5月14日

查看全文
  • Kedar Nagnathrao Ghogale,
  • Ravindra Sadashivrao Apare,
  • Ravindra H. Borhade

摘要

Person Re-identification (Re-ID) is employed for the identification of persons in several cameras. However, the process of detecting the person is not an easy task, because of the lack of database, and distance. Therefore, a new model known as Hybrid DeepID FaceNet (HyDeepIDF-Net) is employed for person Re-ID. At first, input video obtained from the dataset is allowed for frame extraction for the extraction of key frames from the input video. Next, face detection is done by Viola-Jone’s algorithm from the extracted keyframes. Finally, the residual image extraction from the detected face image is performed by employing Deep Residual Network (DRN). Similarly, the acquired face image for query is separately allowed for residual image extraction using DRN. Then, the resultant extracted residual images from query and input images are allowed for semantic matching using the proposed HyDeepIDF-Net. The HyDeepIDF-Net is created by accumulating DeeepID-Net and FaceNet. Here, DeepId Net and FaceNet are employed as they can find the similarities and differences between faces. At last, the tracked persons are annotated on the frames. Further, the assessment of HyDeepIDF-Net in terms of recognition percentage, accuracy, precision, F1-score, and recall show that maximum values of 92.12%, 93.91%, 91.41%, 92.32%, and 93.24%, and minimum latency of 0.54s are recorded.