深度学习跟SLAM的结合点
作者:互联网
1、用深度学习方法替换传统SLAM中的一个/几个模块
- 特征提取,特征匹配,提高特征点稳定性,提取点线面等不同层级的特征点。
- 深度估计
- 位姿估计
- 重定位
- 其他
2、在传统SLAM之上加入语义信息(毕设相关)
- 图像语义分割
- 语义地图构建
3、端到端的SLAM
- 机器人自主导航(深度强化学习)等
2.1 Semantc SLAM
特点:
- ORB_SLAM2, used as our SLAM backend.
- pytorch-semseg, used as our semantic segmantation library.
- octomap, used as our map representation.
- pcl library, used for point cloud processing.
2.2 ORB-SLAM2+YOLO3
Qi X, Yang S, Yan Y. Deep Learning Based Semantic Labelling of 3D Point Cloud in Visual SLAM[C]//IOP Conference Series: Materials Science and Engineering. IOP Publishing, 2018, 428(1): 012023.
国防科技大学高性能计算国家重点实验室
code:orb-slam2_with_semantic_label
2.3 Meaningful maps with object-oriented semantic mappng
Sünderhauf N, Pham T T, Latif Y, et al. Meaningful maps with object-oriented semantic mapping[C]//2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2017: 5079-5085.
ORB-SLAM2+SSD
2.4 DynaSLAM: Tracking, mapping, and inpainting in dynamic scenes
ORB-SLAM2 + Mask R-CNN
code:DynaSLAM
2.5 DA-RNN: Semantic mapping with data associated recurrent neural networks
RNN+CNN 语义分割
2.6 Ds-slam: A semantic visual slam towards dynamic environments
SegNet+ORB SLAM2
2.7 Maskfusion: Real-time recognition, tracking and reconstruction of multiple moving objects
MaskFusion + ElasticFusion
标签:used,semantic,结合点,语义,SLAM2,SLAM,深度,ORB 来源: https://www.cnblogs.com/bupt213/p/12088409.html