Scholarlink

In urban environments, Autonomous Vehicles (AVs) face various challenges, particularly in dealing with occlusions. This work delves into vehicle-to-everything (V2X) cooperation to enhance AVs’ perception in the 3D object detection task. Collaboration combines fields of view from various agents to enable AVs to ’see’ through occlusions and detect objects that are either poorly visible or completely outside their range. This enhancement is crucial to ensure safer and more efficient urban navigation. We specifically focus on the late fusion of LiDAR data streams, using a Breadth-First Search (BFS) algorithm to group bounding boxes with high Intersection-over-Union (IoU). The resulting clusters are then fused via a weighted averaging scheme, producing more accurate and robust 3D detections. This BFS-based weighted method is straightforward, computationally efficient, and requires minimal communication overhead, making it well-suited for real-time urban driving scenarios.

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

BFS-Based Weighted Late Fusion for 3D Object LiDAR Detection in V2X Systems

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Selma Nourhane Oubouabdellah,
  • Elwan Hery,
  • Julien Moreau,
  • Vincent Fremont

摘要

In urban environments, Autonomous Vehicles (AVs) face various challenges, particularly in dealing with occlusions. This work delves into vehicle-to-everything (V2X) cooperation to enhance AVs’ perception in the 3D object detection task. Collaboration combines fields of view from various agents to enable AVs to ’see’ through occlusions and detect objects that are either poorly visible or completely outside their range. This enhancement is crucial to ensure safer and more efficient urban navigation. We specifically focus on the late fusion of LiDAR data streams, using a Breadth-First Search (BFS) algorithm to group bounding boxes with high Intersection-over-Union (IoU). The resulting clusters are then fused via a weighted averaging scheme, producing more accurate and robust 3D detections. This BFS-based weighted method is straightforward, computationally efficient, and requires minimal communication overhead, making it well-suited for real-time urban driving scenarios.