From David Scaramuzza via ros-users@
The code is available at: https://github.com/uzh-rpg/rpg_open_remode
It implements a "REgularized, probabilistic, MOnocular Depth Estimation", as described in the paper:
M. Pizzoli, C. Forster, D. Scaramuzza
REMODE: Probabilistic, monocular dense reconstruction in real time
IEEE International Conference on Robotics and Automation (ICRA), pp. 2609-2616, 2014
The idea is to achieve real-time performance by combining Bayesian, per-pixel estimation with a fast regularization scheme that takes into account the measurement uncertainty to provide spatial regularity and mitigate the effect of noise.
Namely, a probabilistic depth measurement is carried out in real time for each pixel and the computed uncertainty is used to reject erroneous estimations and provide live feedback on the reconstruction progress.
The novelty of the regularization is that the estimated depth uncertainty from the per-pixel depth estimation is used to weight the smoothing.
Since it provides real-time, dense depth maps along with the corresponding confidence maps, REMODE is very suitable for robotic applications, such as environment interaction, motion planning, active vision and control, where both dense information and map uncertainty may be required.
More info here: http://rpg.ifi.uzh.ch/research_dense.html
The open source implementation requires a CUDA capable GPU and the NVIDIA CUDA Toolkit.
Instructions for building and running the code are available in the repository wiki.
We are happy to release an open source implementation of our approach for real-time, monocular, dense depth estimation, called "REMODE".
The code is available at: https://github.com/uzh-rpg/rpg
It implements a "REgularized, probabilistic, MOnocular Depth Estimation", as described in the paper:
M. Pizzoli, C. Forster, D. Scaramuzza
REMODE: Probabilistic, monocular dense reconstruction in real time
IEEE International Conference on Robotics and Automation (ICRA), pp. 2609-2616, 2014
The idea is to achieve real-time performance by combining Bayesian, per-pixel estimation with a fast regularization scheme that takes into account the measurement uncertainty to provide spatial regularity and mitigate the effect of noise.
Namely, a probabilistic depth measurement is carried out in real time for each pixel and the computed uncertainty is used to reject erroneous estimations and provide live feedback on the reconstruction progress.
The novelty of the regularization is that the estimated depth uncertainty from the per-pixel depth estimation is used to weight the smoothing.
Since it provides real-time, dense depth maps along with the corresponding confidence maps, REMODE is very suitable for robotic applications, such as environment interaction, motion planning, active vision and control, where both dense information and map uncertainty may be required.
More info here: http://rpg.ifi.uzh.ch/research
The open source implementation requires a CUDA capable GPU and the NVIDIA CUDA Toolkit.
Instructions for building and running the code are available in the repository wiki.
Leave a comment