​单目视觉三位重建的思路方法

2018-03-22 10:31:09 40

单目视觉三位重建的思路方法

 1. Optical Flow/Normal Flow --> Motion Estimation --> Depth 
  • 从Normal Flow估计运动参数:D Yuan[2] 有数篇文章发在SCI上(这里给出最新的一篇)

  • 从光流估计运动参数:方法不少,比如[4]

2. SLAM:
3. SfM:
  • SfM我了解的不多,但感觉和SLAM非常相近

  • MLM-SFM[5]: 做车辆自动驾驶相关的,前向运动、车辆探测、距离估计,还是实时的…

  • LIBVISO(Andreas Geiger): 视频youtu.be/EPTJz7w_AqU(这个不知道该归到哪一类,暂且放在SfM里吧)

4. Learning Based:
  • 这类方法应该是用单目相机从单幅图像中估计深度,从一开始的半自动(需要添加特征点、纹理之类的)到现在的全自动,现在用学习的方法越来越多了,譬如:[6]

5. 有个DataSet:The KITTI Vision Benchmark Suite(可以做光流、里程计、vSLAM之类的)

欢迎讨论批评指正,想到了新的东西再更新。

Reference
[1] Davison, Andrew J., et al. "MonoSLAM: Real-time single camera SLAM."Pattern Analysis and Machine Intelligence, IEEE Transactions on 29.6 (2007): 1052-1067.
[2] Engel, Jakob, Thomas Schöps, and Daniel Cremers. "LSD-SLAM: Large-scale direct monocular SLAM." Computer Vision–ECCV 2014. Springer International Publishing, 2014. 834-849.
[3] Yuan, Ding, et al. "Camera motion estimation through monocular normal flow vectors." Pattern Recognition Letters 52 (2015): 59-64.
[4] Raudies, Florian, and Heiko Neumann. "An efficient linear method for the estimation of ego-motion from optical flow." Pattern Recognition. Springer Berlin Heidelberg, 2009. 11-20.
[5] Song, Shiyu, and Manmohan Chandraker. "Robust Scale Estimation in Real-Time Monocular SFM for Autonomous Driving." Computer Vision and Pattern Recognition (CVPR), 2014 IEEE Conference on. IEEE, 2014.
[6] Karsch, Kevin, Ce Liu, and S. Kang. "Depthtransfer: Depth extraction from video using non-parametric sampling." (2014): 1-1.