{"title":"改进的激光雷达点云法向量引导语义分割网络","authors":"Minsung Kim, Inyoung Oh, Dongho Yun, Kwanghee Ko","doi":"10.1093/jcde/qwad102","DOIUrl":null,"url":null,"abstract":"Abstract As LiDAR sensors become increasingly prevalent in the field of autonomous driving, the need for accurate semantic segmentation of 3D points grows accordingly. To address this challenge, we propose a novel network model that enhances segmentation performance by utilizing normal vector information. Firstly, we present a method to improve the accuracy of normal estimation by using the intensity and reflection angles of the light emitted from the LiDAR sensor. Secondly, we introduce a novel local feature aggregation module that integrates normal vector information into the network to improve the performance of local feature extraction. The normal information is closely related to the local structure of the shape of an object, which helps the network to associate unique features with corresponding objects. We propose four different structures for local feature aggregation, evaluate them, and choose the one that shows the best performance. Experiments using the SemanticKITTI dataset demonstrate that the proposed architecture outperforms both the baseline model, RandLA-Net, and other existing methods, achieving mean Intersection over Union (mIoU) of 57.9%. Furthermore, it shows highly competitive performance compared to RandLA-Net for small and dynamic objects in a real road environment. For example, it yielded 95.2% for cars, 47.4% for bicycles, 41.0% for motorcycles, 57.4% for bicycles, and 53.2% for pedestrians.","PeriodicalId":48611,"journal":{"name":"Journal of Computational Design and Engineering","volume":"132 16","pages":"0"},"PeriodicalIF":4.8000,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Improved semantic segmentation network using normal vector guidance for LiDAR point clouds\",\"authors\":\"Minsung Kim, Inyoung Oh, Dongho Yun, Kwanghee Ko\",\"doi\":\"10.1093/jcde/qwad102\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Abstract As LiDAR sensors become increasingly prevalent in the field of autonomous driving, the need for accurate semantic segmentation of 3D points grows accordingly. To address this challenge, we propose a novel network model that enhances segmentation performance by utilizing normal vector information. Firstly, we present a method to improve the accuracy of normal estimation by using the intensity and reflection angles of the light emitted from the LiDAR sensor. Secondly, we introduce a novel local feature aggregation module that integrates normal vector information into the network to improve the performance of local feature extraction. The normal information is closely related to the local structure of the shape of an object, which helps the network to associate unique features with corresponding objects. We propose four different structures for local feature aggregation, evaluate them, and choose the one that shows the best performance. Experiments using the SemanticKITTI dataset demonstrate that the proposed architecture outperforms both the baseline model, RandLA-Net, and other existing methods, achieving mean Intersection over Union (mIoU) of 57.9%. Furthermore, it shows highly competitive performance compared to RandLA-Net for small and dynamic objects in a real road environment. For example, it yielded 95.2% for cars, 47.4% for bicycles, 41.0% for motorcycles, 57.4% for bicycles, and 53.2% for pedestrians.\",\"PeriodicalId\":48611,\"journal\":{\"name\":\"Journal of Computational Design and Engineering\",\"volume\":\"132 16\",\"pages\":\"0\"},\"PeriodicalIF\":4.8000,\"publicationDate\":\"2023-11-13\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Journal of Computational Design and Engineering\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1093/jcde/qwad102\",\"RegionNum\":2,\"RegionCategory\":\"工程技术\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Computational Design and Engineering","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1093/jcde/qwad102","RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS","Score":null,"Total":0}
Improved semantic segmentation network using normal vector guidance for LiDAR point clouds
Abstract As LiDAR sensors become increasingly prevalent in the field of autonomous driving, the need for accurate semantic segmentation of 3D points grows accordingly. To address this challenge, we propose a novel network model that enhances segmentation performance by utilizing normal vector information. Firstly, we present a method to improve the accuracy of normal estimation by using the intensity and reflection angles of the light emitted from the LiDAR sensor. Secondly, we introduce a novel local feature aggregation module that integrates normal vector information into the network to improve the performance of local feature extraction. The normal information is closely related to the local structure of the shape of an object, which helps the network to associate unique features with corresponding objects. We propose four different structures for local feature aggregation, evaluate them, and choose the one that shows the best performance. Experiments using the SemanticKITTI dataset demonstrate that the proposed architecture outperforms both the baseline model, RandLA-Net, and other existing methods, achieving mean Intersection over Union (mIoU) of 57.9%. Furthermore, it shows highly competitive performance compared to RandLA-Net for small and dynamic objects in a real road environment. For example, it yielded 95.2% for cars, 47.4% for bicycles, 41.0% for motorcycles, 57.4% for bicycles, and 53.2% for pedestrians.
期刊介绍:
Journal of Computational Design and Engineering is an international journal that aims to provide academia and industry with a venue for rapid publication of research papers reporting innovative computational methods and applications to achieve a major breakthrough, practical improvements, and bold new research directions within a wide range of design and engineering:
• Theory and its progress in computational advancement for design and engineering
• Development of computational framework to support large scale design and engineering
• Interaction issues among human, designed artifacts, and systems
• Knowledge-intensive technologies for intelligent and sustainable systems
• Emerging technology and convergence of technology fields presented with convincing design examples
• Educational issues for academia, practitioners, and future generation
• Proposal on new research directions as well as survey and retrospectives on mature field.