Scholarlink

To address the challenge of recognizing occluded aerial targets, this paper proposes a ground-based multi-view recognition scheme. This system maintains accurate target perception even when some robot sensors fail to provide effective information. By employing an evaluation network-based multi-view data fusion method, the scheme flexibly utilizes the complementarity of sensing data from multiple robots. The evaluation network mechanism serves to determine the completeness of valid information in images from different perspectives. When targets are in occlusion scenarios where critical information is severely lost in certain images, the scoring mechanism significantly reduces the weight of these images in the overall recognition task. Experimental results demonstrate that the proposed ground-based multi-view recognition algorithm achieves an average improvement of over 16% in recognition effectiveness compared to aerial recognition methods.

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

A Occlusion Target Detection Technique Based on Multi-view Fusion

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Junqin Lin,
  • Yanjiang Chen,
  • Kui Huang,
  • Zhiyuan Yu

摘要

To address the challenge of recognizing occluded aerial targets, this paper proposes a ground-based multi-view recognition scheme. This system maintains accurate target perception even when some robot sensors fail to provide effective information. By employing an evaluation network-based multi-view data fusion method, the scheme flexibly utilizes the complementarity of sensing data from multiple robots. The evaluation network mechanism serves to determine the completeness of valid information in images from different perspectives. When targets are in occlusion scenarios where critical information is severely lost in certain images, the scoring mechanism significantly reduces the weight of these images in the overall recognition task. Experimental results demonstrate that the proposed ground-based multi-view recognition algorithm achieves an average improvement of over 16% in recognition effectiveness compared to aerial recognition methods.