이미지 프로세싱 모폴로지 연산 및 딥러닝을 활용한 3차원 점군지도의동적 물체 제거 방법

Abstract

The autonomous navigation system requires a static map for better usability. However, urban environments are filled with dynamic objects, such as vehicles and pedestrians that leave traces in the 3D point cloud map. These traces can act as obstacles and impede navigation systems. As a result, extensive studies have been conducted to remove dynamic objects based on geometric information, so it is challenging to remove dynamic objects in highly dynamic environments. To tackle this problem, we propose a method for dynamic object removal in 3D point cloud maps. The proposed method combines a pseudo occupancy-based method, morphological image processing, and deep learning.