MosViT: towards vision transformers for moving object segmentation based on Lidar point cloud

Author:

Ma ChunyunORCID,Shi Xiaojun,Wang Yingxin,Song Shuai,Pan Zhen,Hu Jiaxiang

Abstract

Abstract Moving object segmentation is fundamental for various downstream tasks in robotics and autonomous driving, providing crucial information for them. Effectively extracting spatial-temporal information from consecutive frames and addressing the scarcity of dataset is important for learning-based 3D LiDAR moving object segmentation (LIDAR-MOS). In this work, we propose a novel deep neural network based on vision transformers (ViTs) to tackle this problem. We first validate the feasibility of transformer networks for this task, offering an alternative to CNNs. Specifically, we utilize a dual-branch structure using range (residual) image as input to extract spatial-temporal information from consecutive frames and fuse it using a motion-guided attention mechanism. Furthermore, we employ the ViT as the backbone, keeping its architecture unchanged from what is used for RGB images. This enables us to leverage pre-trained models on RGB images to improve results, addressing the issue of limited LiDAR point cloud data, which is cheaper compared to acquiring and annotating point cloud data. We validate the effectiveness of our approach on the LIDAR-MOS benchmark of SemanticKitti and achieve comparable results to methods that use CNNs on range image data. The source code and trained models will be available at https://github.com/mafangniu/MOSViT.git.

Funder

Research and Development Program of China

Publisher

IOP Publishing

Reference51 articles.

1. Small obstacle avoidance based on RGB-D semantic segmentation;Hua,2019

2. Multi-robot mission planning in dynamic semantic environments;Samarth,2023

3. Suma++: efficient lidar-based semantic slam;Chen,2019

4. Efficient spatial-temporal information fusion for lidar-based 3d moving object segmentation;Sun,2022

5. Deep learning for 3d point clouds: a survey;Guo;IEEE Trans. Pattern Anal. Mach. Intell.,2020

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3