스테레오 RGB-E를 이용한 동적 환경에서 강인한 SLAM 방법
Robust SLAM Method in Dynamic Environments Using Stereo RGB-E
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Standard camera based Simultaneous Localization and Mapping (SLAM) has been combined with deep learning-based semantic segmentation networks to achieve robust performance in dynamic environments. However, these SLAM methods heavily depend on the performance of the semantic segmentation network, which in turn relies on the quality of the input images. When input images are degraded by motion blur or other factors, the performance of SLAM also declines. To address this issue, we propose a method that integrates an event camera-based semantic segmentation network with an image-based SLAM algorithm. Event cameras, which asynchronously capture brightness changes, offer high temporal resolution and a wide dynamic range, making them advantageous in complex environments. The proposed method uses event data to predict semantic masks, removing dynamic objects and enabling more accurate localization. Evaluations on an urban dataset demonstrate the method’s robust performance in dynamic environments, highlighting its practical potential.
Abstract
Standard camera based Simultaneous Localization and Mapping (SLAM) has been combined with deep learning-based semantic segmentation networks to achieve robust performance in dynamic environments. However, these SLAM methods heavily depend on the performance of the semantic segmentation network, which in turn relies on the quality of the input images. When input images are degraded by motion blur or other factors, the performance of SLAM also declines. To address this issue, we propose a method that integrates an event camera-based semantic segmentation network with an image-based SLAM algorithm. Event cameras, which asynchronously capture brightness changes, offer high temporal resolution and a wide dynamic range, making them advantageous in complex environments. The proposed method uses event data to predict semantic masks, removing dynamic objects and enabling more accurate localization. Evaluations on an urban dataset demonstrate the method’s robust performance in dynamic environments, highlighting its practical potential.