Scholarlink

This paper presents an cost-effective approach for geo-localization and location-aware object detection using a single 360 \(^{\circ }\) fisheye camera lens on mobile platforms such as street cleaning vehicles. We propose a system that captures a comprehensive view of the surroundings and accurately detects people and objects. Using the camera’s geometry, the system infers distances to objects on the ground and projects them into global coordinates, creating a temporal spatial map. This ‘memory map’ is continuously updated, allowing for the accumulation of detection predictions over time. This approach significantly enhances the robustness and accuracy of object detection in dynamic environments. Our experiments demonstrate the system’s efficacy, making it a strong candidate for implementation in various real-world applications requiring enhanced situational awareness and autonomous decision-making capabilities.

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

Robust Single-Cam Surround View Object Detection and Localization Using Memory Maps

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yitong Quan,
  • Benjamin Kiefer,
  • Martin Messmer,
  • Charan Ram Akupati,
  • Rainer Graser,
  • Andreas Zell

摘要

This paper presents an cost-effective approach for geo-localization and location-aware object detection using a single 360 \(^{\circ }\) fisheye camera lens on mobile platforms such as street cleaning vehicles. We propose a system that captures a comprehensive view of the surroundings and accurately detects people and objects. Using the camera’s geometry, the system infers distances to objects on the ground and projects them into global coordinates, creating a temporal spatial map. This ‘memory map’ is continuously updated, allowing for the accumulation of detection predictions over time. This approach significantly enhances the robustness and accuracy of object detection in dynamic environments. Our experiments demonstrate the system’s efficacy, making it a strong candidate for implementation in various real-world applications requiring enhanced situational awareness and autonomous decision-making capabilities.