Scholarlink

The management of urban roadside tree positioning is a crucial aspect of modern urban governance. Traditionally, manual surveys have been employed to determine the locations of trees, which is both time-consuming and labor-intensive. Recently, the use of high-resolution satellite imagery and object detection algorithms for roadside tree localization has emerged as a new trend, enabling large-scale automated recognition. However, accurate identification of individual trees remains challenging due to issues like overlapping canopies. In contrast, street view imagery provides a clear ground-level perspective of individual trees, but its frontal view limitation hinders the determination of actual geographic coordinates and integration with satellite imagery results. This study introduces a ray-casting Orientation Mapping method, which determines the actual location of trees by intersecting the azimuth angle of trees in street view images with pedestrian paths on OpenStreetMap (OSM). This method allows for the integration of street view and satellite imagery results under a top-down perspective. By combining multi-source data (satellite, street view, OSM), the tree localization method proposed in this paper demonstrates a 24% improvement in prediction accuracy compared to using satellite imagery alone.

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

Urban Street Tree Recognition Based on Deep Learning and Multi-Source Data

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Xinjie Huo,
  • Hanxi Wang,
  • Sirui Chen,
  • Junyan Ye,
  • Weijia Li

摘要

The management of urban roadside tree positioning is a crucial aspect of modern urban governance. Traditionally, manual surveys have been employed to determine the locations of trees, which is both time-consuming and labor-intensive. Recently, the use of high-resolution satellite imagery and object detection algorithms for roadside tree localization has emerged as a new trend, enabling large-scale automated recognition. However, accurate identification of individual trees remains challenging due to issues like overlapping canopies. In contrast, street view imagery provides a clear ground-level perspective of individual trees, but its frontal view limitation hinders the determination of actual geographic coordinates and integration with satellite imagery results. This study introduces a ray-casting Orientation Mapping method, which determines the actual location of trees by intersecting the azimuth angle of trees in street view images with pedestrian paths on OpenStreetMap (OSM). This method allows for the integration of street view and satellite imagery results under a top-down perspective. By combining multi-source data (satellite, street view, OSM), the tree localization method proposed in this paper demonstrates a 24% improvement in prediction accuracy compared to using satellite imagery alone.