Scholarlink

<p>Thermal images are predominantly used in surveillance and search and rescue missions especially when visibility is limited due to factors like darkness, fog, rain, or occlusions. Person detection in thermal images face several challenges such as low contrast, noise, and variations in temperature due to environmental conditions. In addition to above challenges the other factors such as lack of color information, occlusion will make the person detection more difficult. Detecting persons in certain conditions remains a difficult task. To challenge these complications, the paper integrates histogram-based image enhancement to improve visibility under challenging conditions and utilizes the YOLOv8 model for person detection. For enhancing the image can be enhanced&#xa0;through kurtosis-based histogram equalization approach and later the enhanced images are applied to YOLOv8 model for detecting people. The proposed method is tested on various datasets such as OSU thermal pedestrian dataset, FLIR ADAS Dataset, and AAU PD T dataset. The proposed method has been compared qualitatively and quantitatively using the parameters F1 score, Accuracy, precision and mean Average Precision (mAP) by the other standard histogram enhancement methods. By using the kurtosis-based histogram enhancement approach with YOLOv8 shows superior value of mAP of 0.99 for OSU thermal dataset and 0.9 for AAU PD T dataset and 0.93 for FLIR ADAS dataset when compared with other methods.</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

Person detection in thermal images using kurtosis based histogram enhancement and YOLOv8

  • 期刊论文

  • 发表日期：

    2025年4月21日

查看全文
  • Tirupathamma Mudavath,
  • Vandana Niranjan

摘要

Thermal images are predominantly used in surveillance and search and rescue missions especially when visibility is limited due to factors like darkness, fog, rain, or occlusions. Person detection in thermal images face several challenges such as low contrast, noise, and variations in temperature due to environmental conditions. In addition to above challenges the other factors such as lack of color information, occlusion will make the person detection more difficult. Detecting persons in certain conditions remains a difficult task. To challenge these complications, the paper integrates histogram-based image enhancement to improve visibility under challenging conditions and utilizes the YOLOv8 model for person detection. For enhancing the image can be enhanced through kurtosis-based histogram equalization approach and later the enhanced images are applied to YOLOv8 model for detecting people. The proposed method is tested on various datasets such as OSU thermal pedestrian dataset, FLIR ADAS Dataset, and AAU PD T dataset. The proposed method has been compared qualitatively and quantitatively using the parameters F1 score, Accuracy, precision and mean Average Precision (mAP) by the other standard histogram enhancement methods. By using the kurtosis-based histogram enhancement approach with YOLOv8 shows superior value of mAP of 0.99 for OSU thermal dataset and 0.9 for AAU PD T dataset and 0.93 for FLIR ADAS dataset when compared with other methods.