Scholarlink

<p>Indoor layout estimation is a fundamental task in spatial computing, providing essential structural information for VR and AR applications, including scene understanding, spatial mapping, and navigation. Existing RGB-based approaches struggle with highly dynamic head movements in XR scenarios, leading to motion blur and degraded estimation accuracy. To address these challenges, we propose an event-based layout estimation framework designed for dynamic XR applications. Unlike conventional methods, <i>DynoLayout</i> leverages a Spatiotemporal Parallel Adaptive Representation (SPAR) to encode both spatial correlations and temporal dependencies in event-streams. Furthermore, we introduce a redundant line merging module that refines layout predictions by analyzing spatial correlations and line-adjacent event appearance distributions (LEAD), reducing the redundant estimations. We evaluate <i>DynoLayout</i> on the EV-Layout dataset, demonstrating its robustness across various event-stream lengths and its superiority over existing event-based representations. Our results highlight the effectiveness of <i>DynoLayout</i> in achieving accurate and reliable layout estimation under highly dynamic XR 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

Dynolayout: robust layout estimation from event-stream for extended reality under dynamic scenarios

  • 期刊论文

  • 发表日期：

    2025年12月9日

查看全文
  • Bing Li,
  • Xucheng Guo,
  • Yaqi Zhao,
  • Qiang Qu,
  • Guangrong Zhao,
  • Yuanfeng Zhou,
  • Yiran Shen

摘要

Indoor layout estimation is a fundamental task in spatial computing, providing essential structural information for VR and AR applications, including scene understanding, spatial mapping, and navigation. Existing RGB-based approaches struggle with highly dynamic head movements in XR scenarios, leading to motion blur and degraded estimation accuracy. To address these challenges, we propose an event-based layout estimation framework designed for dynamic XR applications. Unlike conventional methods, DynoLayout leverages a Spatiotemporal Parallel Adaptive Representation (SPAR) to encode both spatial correlations and temporal dependencies in event-streams. Furthermore, we introduce a redundant line merging module that refines layout predictions by analyzing spatial correlations and line-adjacent event appearance distributions (LEAD), reducing the redundant estimations. We evaluate DynoLayout on the EV-Layout dataset, demonstrating its robustness across various event-stream lengths and its superiority over existing event-based representations. Our results highlight the effectiveness of DynoLayout in achieving accurate and reliable layout estimation under highly dynamic XR conditions.