Scholarlink

Accurate and reliable location services are the fundamental key technologies for autonomous driving and intelligent transportation. Due to the strong positioning complementarity between GNSS and visual positioning, GNSS/vision combined system has been widely used in navigation and positioning fields such as robots. This research presents a loosely coupled GNSS/visual localization approach that integrates environmental semantics to mitigate the degradation of positioning accuracy in GNSS/ visual integrated systems operating in complex urban environments. Firstly, the improved YOLOv5 model was used to extract the environment semantic information in the image, and the dynamic feature points that affected the visual positioning accuracy were eliminated. Then, the occlusion of GNSS signals by buildings was evaluated through environmental semantics, and an adaptive Kalman filter model based on occlusion factors was constructed to improve the positioning accuracy of the GNSS/ vision combined system. Finally, the on-board experimental results showed that the proposed method has smaller Root Mean Square Error (RMSE) and achieves 10.7% and 9.4% higher positioning accuracy in the east and north directions, respectively, compared to a GNSS/INS/vision tightly coupled system.

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

GNSS/ Visual Fusion Positioning in Urban Environment Based on Semantic Segmentation and Adaptive Filtering

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Li Xin,
  • Li Jian,
  • Jia Ruixin,
  • Zhang Wanjuan,
  • Cheng Junbing

摘要

Accurate and reliable location services are the fundamental key technologies for autonomous driving and intelligent transportation. Due to the strong positioning complementarity between GNSS and visual positioning, GNSS/vision combined system has been widely used in navigation and positioning fields such as robots. This research presents a loosely coupled GNSS/visual localization approach that integrates environmental semantics to mitigate the degradation of positioning accuracy in GNSS/ visual integrated systems operating in complex urban environments. Firstly, the improved YOLOv5 model was used to extract the environment semantic information in the image, and the dynamic feature points that affected the visual positioning accuracy were eliminated. Then, the occlusion of GNSS signals by buildings was evaluated through environmental semantics, and an adaptive Kalman filter model based on occlusion factors was constructed to improve the positioning accuracy of the GNSS/ vision combined system. Finally, the on-board experimental results showed that the proposed method has smaller Root Mean Square Error (RMSE) and achieves 10.7% and 9.4% higher positioning accuracy in the east and north directions, respectively, compared to a GNSS/INS/vision tightly coupled system.