Scholarlink

<p>Approximately 23% of Rio de Janeiro’s population, around 1.5 million people, live in <i>favelas</i>, a form of informal settlement in Brazil. These communities often lack essential infrastructure and are disproportionately impacted by environmental and public health challenges linked to their complex and irregular urban morphology. Given their scale and significance, a comprehensive understanding of the spatial configuration of these communities is critical. Yet, conventional top-down mapping methods fail to adequately capture the vertical and spatial intricacies within these environments. This research proposes a novel framework that incorporates a semantic segmentation procedure for three-dimensional (3D) scene understanding of favelas using a wearable Light Detection and Ranging (LiDAR) dataset and an artificial intelligence (AI)-driven approach. High-resolution point clouds collected in the Vidigal favela were used to train a semantic segmentation model that accurately classifies seven key object categories relevant to the 3D morphology of informal settlements, including ground, vegetation, building, wire, rock, pole, and movable objects. Our results highlight the effectiveness of point level feature engineering and multiscale neighborhood analysis for accurate segmentation of objects in dense and irregular informal environments. This approach provides a scalable way to capture the internal spatial complexity of informal settlements, supporting data-driven planning, risk assessment, and equitable infrastructure development in otherwise inaccessible areas.</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

Unveiling the internal structures of informal settlements through AI-driven automated segmentation of LiDAR data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月10日

查看全文
  • Chang Liu,
  • Kee Moon Jang,
  • Stefania Dimitrov,
  • Cezar Lima Ferreira Barbalho,
  • Fabio Duarte,
  • Carlo Ratti

摘要

Approximately 23% of Rio de Janeiro’s population, around 1.5 million people, live in favelas, a form of informal settlement in Brazil. These communities often lack essential infrastructure and are disproportionately impacted by environmental and public health challenges linked to their complex and irregular urban morphology. Given their scale and significance, a comprehensive understanding of the spatial configuration of these communities is critical. Yet, conventional top-down mapping methods fail to adequately capture the vertical and spatial intricacies within these environments. This research proposes a novel framework that incorporates a semantic segmentation procedure for three-dimensional (3D) scene understanding of favelas using a wearable Light Detection and Ranging (LiDAR) dataset and an artificial intelligence (AI)-driven approach. High-resolution point clouds collected in the Vidigal favela were used to train a semantic segmentation model that accurately classifies seven key object categories relevant to the 3D morphology of informal settlements, including ground, vegetation, building, wire, rock, pole, and movable objects. Our results highlight the effectiveness of point level feature engineering and multiscale neighborhood analysis for accurate segmentation of objects in dense and irregular informal environments. This approach provides a scalable way to capture the internal spatial complexity of informal settlements, supporting data-driven planning, risk assessment, and equitable infrastructure development in otherwise inaccessible areas.