基本信息
浏览量:11
![](https://originalfileserver.aminer.cn/sys/aminer/icon/show-trajectory.png)
个人简介
We present a deep learning system that learns to infer a scene’s dense depth map from sparse depth points and images of indoor scenes. Specifically, because a visual-inertial simultaneous localization and mapping (VI-SLAM) system only provide depth values for a small percentage of the pixels which lie mainly over the high-textured areas (sparse depth), we compensate the missing depth of the low textured ones by leveraging their planar structures and surface normals which is an important intermediate representation, then use Deep Convolutional Neural Network (CNN) to complete and obtain the depth for each pixel in the image (dense depth). We show that our method significantly improves the baseline and also outperforms other state-of-the-art approaches both on training (ScanNet and NYUv2) and testing (collected with Azure Kinect) datasets. Finally, we hope this work can serve as an effective baseline in leveraging semantic information to improve dense depth estimation from VI-SLAM which has important and direct applications in augmented reality, robot motion planning, and 3D mapping.
研究兴趣
论文共 7 篇作者统计合作学者相似作者
按年份排序按引用量排序主题筛选期刊级别筛选合作者筛选合作机构筛选
时间
引用量
主题
期刊级别
合作者
合作机构
IEEE Conference on Computer Vision and Pattern Recognitionno. 1 (2022): 2822-2831
作者统计
合作学者
合作机构
D-Core
- 合作者
- 学生
- 导师
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn