Scholarlink

<p>Ground-based unmanned systems have witnessed rapid global development; however, limitations in sensor field of view and resolution often hinder accurate target recognition and localization. This study proposes an integrated approach leveraging binocular cameras and infrared thermal imaging technology, specifically focusing on human target detection. An experimental binocular vision platform was constructed, and an improved SSD target detection model was developed. This includes an enhanced HOG detection layer to refine edge feature extraction, a lightweight MobileNet structure to accelerate detection speed, and a CBAM attention module to boost detection accuracy. Additionally, the ORB matching algorithm was optimized using the GMS method to enhance binocular image alignment, thereby improving target localization precision. Experimental results demonstrated a 5.5% increase in human target detection accuracy and a 62% boost in detection speed compared to the original SSD model. While localization error increases with depth, it remains below 1% within 3&#xa0;m, confirming the method’s reliability. This study advances the field by providing a robust and efficient framework for target recognition and localization in complex environments.</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

Enhanced target recognition and localization using binocular vision and infrared thermal imaging

  • 期刊论文

  • 发表日期：

    2025年8月23日

查看全文
  • Chunjian Su,
  • Luhui Li,
  • Hongen Wei,
  • Hening Sun,
  • Yongxu Chen,
  • Daolong Zhang,
  • Tinyi Din,
  • Chenming Li

摘要

Ground-based unmanned systems have witnessed rapid global development; however, limitations in sensor field of view and resolution often hinder accurate target recognition and localization. This study proposes an integrated approach leveraging binocular cameras and infrared thermal imaging technology, specifically focusing on human target detection. An experimental binocular vision platform was constructed, and an improved SSD target detection model was developed. This includes an enhanced HOG detection layer to refine edge feature extraction, a lightweight MobileNet structure to accelerate detection speed, and a CBAM attention module to boost detection accuracy. Additionally, the ORB matching algorithm was optimized using the GMS method to enhance binocular image alignment, thereby improving target localization precision. Experimental results demonstrated a 5.5% increase in human target detection accuracy and a 62% boost in detection speed compared to the original SSD model. While localization error increases with depth, it remains below 1% within 3 m, confirming the method’s reliability. This study advances the field by providing a robust and efficient framework for target recognition and localization in complex environments.