Small Defect Instance Reconstruction Based on 2D Connectivity-3D Probabilistic Voting

CORC > 沈阳自动化研究所 > 中国科学院沈阳自动化研究所 > 工艺装备与智能机器人研究室

	Small Defect Instance Reconstruction Based on 2D Connectivity-3D Probabilistic Voting
	Hong KL(洪坤龙)2,3,4; Wang HG(王洪光)3,4; Zhu, Bing 1
	2021
会议日期	December 27-31, 2021
会议地点	Sanya, China
页码	1448-1453
英文摘要	The detection and statistics of defects are an essential part of monitoring large-scale concrete wall defects. Although CNN (convolution neural network) has achieved high accuracy on defects segmentation, imbalanced categories' (cracks) segment results are not meticulous. Besides, the sparseness of small defect data results in inconsistency in the defect statistics of the global 3D semantic model. This paper applies the improved boundary loss function to the defect segmentation network to enhance the IoU (intersection over union) of small defects segmentation results. Combine the 2D image connectivity and the 3D probabilistic voxel voting mechanism with the TSDF model to achieve minor defects' consistent semantic reconstruction. Experiments show that rectified boundary loss can segment cracks more fastidious. The threshold probabilistic voting method has augmented the consistency of small defects in the large 3D model.
源文献作者	Chiba Institute of Technology ; et al. ; IEEE Robotics and Automation Society ; Nankai University ; Shenyang Institute of Automation ; Shenzhen Academy of Robotics
产权排序	1
会议录	2021 IEEE International Conference on Robotics and Biomimetics, ROBIO 2021
会议录出版者	IEEE
会议录出版地	New　York
语种	英语
ISBN号	978-1-6654-0535-5
内容类型	会议论文
源URL	[http://ir.sia.cn/handle/173321/30838]
专题	工艺装备与智能机器人研究室
通讯作者	Hong KL(洪坤龙)
作者单位	1.China Yangtze Power Co. Ltd., Three Gorges Power Plant, Yichang 443133, China 2.University of Chinese Academy of Sciences, Beijing 100049, China 3.Institutes for Robotics and Intelligent Manufacturing, Chinese Academy of Sciences, Shenyang 110016, China 4.State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang 110016, China
推荐引用方式 GB/T 7714	Hong KL,Wang HG,Zhu, Bing. Small Defect Instance Reconstruction Based on 2D Connectivity-3D Probabilistic Voting[C]. 见:. Sanya, China. December 27-31, 2021.