Scholarlink

<p>Aerial images are currently used in a variety of object-tracking applications, including traffic monitoring,&#xa0;disaster prevention, wildlife surveillance, as well as crowd control. Several DL&#xa0;approaches had a significant impact on object recognition as well as tracking. Pre-trained networks are used in DL-based object detectors. If there is an inconsistency between the pre-trained network sector along with the target field, encounters many issues. Because of camera perspective deviation, elevation limits, as well as camera progress, aerial images vary from those employed by pre-trained networks. In this work, the object detection and tracking framework for Aerial images is proposed. The super-pixel motion detection technique employed in this work transforms the input video into frames to obtain the ROI of an image. Then, proposes the CSSMP-ResNet50 model extracts the features like corners, blobs, edges and texture from video frames and then fed these extracted features to further step. Subsequently, proposes an MST-GCRDAN-based detector system that detects the moving objects, and hence the remaining objects are considered as non-moving objects. The proposed MST-GCRDAN-based detector system is a combination of the GCDA network and ROA. The ROA is employed in this work, to optimize the hyper-parameter (weights) of the GCDA network. Consequently, proposes a Kalman filter-based tracking approach that tracks similar moving objects in each frame of single-video segments. At last, the performance of the proposed SI-ROA model is validated with extant methods in terms of certain measures like IOU, NIOU, MOTP, MAP, MOTA, IDFN, IDFP and HOTA respectively. And, the performance (i.e., IOU) of the proposed SI-ROA model attains 81.74% for video 1, 87.24% for video 2, 88.14% for video 3, 86.59% for video 4 and 82.08% for video 5 respectively.</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

RETRACTED ARTICLE: Enhanced graph convolutional remora dual-attentive network-based object detection and tracking framework for aerial images

  • 期刊论文

  • 发表日期：

    2025年3月25日

查看全文
  • Agnivesh Pandey,
  • Rohit Raja

摘要

Aerial images are currently used in a variety of object-tracking applications, including traffic monitoring, disaster prevention, wildlife surveillance, as well as crowd control. Several DL approaches had a significant impact on object recognition as well as tracking. Pre-trained networks are used in DL-based object detectors. If there is an inconsistency between the pre-trained network sector along with the target field, encounters many issues. Because of camera perspective deviation, elevation limits, as well as camera progress, aerial images vary from those employed by pre-trained networks. In this work, the object detection and tracking framework for Aerial images is proposed. The super-pixel motion detection technique employed in this work transforms the input video into frames to obtain the ROI of an image. Then, proposes the CSSMP-ResNet50 model extracts the features like corners, blobs, edges and texture from video frames and then fed these extracted features to further step. Subsequently, proposes an MST-GCRDAN-based detector system that detects the moving objects, and hence the remaining objects are considered as non-moving objects. The proposed MST-GCRDAN-based detector system is a combination of the GCDA network and ROA. The ROA is employed in this work, to optimize the hyper-parameter (weights) of the GCDA network. Consequently, proposes a Kalman filter-based tracking approach that tracks similar moving objects in each frame of single-video segments. At last, the performance of the proposed SI-ROA model is validated with extant methods in terms of certain measures like IOU, NIOU, MOTP, MAP, MOTA, IDFN, IDFP and HOTA respectively. And, the performance (i.e., IOU) of the proposed SI-ROA model attains 81.74% for video 1, 87.24% for video 2, 88.14% for video 3, 86.59% for video 4 and 82.08% for video 5 respectively.