Scholarlink

<p>This paper presents a novel excavator pose estimation framework based on coordinate classification. The framework introduces three key innovations: 1) a coordinate classification network for accurate arm joint estimation under unobstructed conditions, 2) a Kalman filter for maintaining temporal coherence during occlusions, and 3) virtual trajectory-driven recalibration of Kalman parameters for stable pose recovery post-occlusion. The proposed method achieves a lightweight design with a reduction of approximately 80.8% in the number of parameters, while also decreasing the endpoint error (EPE) by 2.81 pixels. Experimental validation on an excavator dataset highlights the robustness and efficiency of the approach, demonstrating its potential for real-time deployment in challenging operational conditions.</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

Excavator pose estimation under occlusion: a coordinate classification approach enhanced by Kalman filtering

  • 期刊论文

  • 发表日期：

    2025年11月7日

查看全文
  • Jian Li,
  • Weili Ding,
  • Guoqing Liu,
  • Changchun Hua,
  • Dengsheng Cai,
  • Jinquan Sun

摘要

This paper presents a novel excavator pose estimation framework based on coordinate classification. The framework introduces three key innovations: 1) a coordinate classification network for accurate arm joint estimation under unobstructed conditions, 2) a Kalman filter for maintaining temporal coherence during occlusions, and 3) virtual trajectory-driven recalibration of Kalman parameters for stable pose recovery post-occlusion. The proposed method achieves a lightweight design with a reduction of approximately 80.8% in the number of parameters, while also decreasing the endpoint error (EPE) by 2.81 pixels. Experimental validation on an excavator dataset highlights the robustness and efficiency of the approach, demonstrating its potential for real-time deployment in challenging operational conditions.