For more information see https://frobelbest.github.io/gslam
- GSLAM: Initialization-robust Monocular Visual SLAM via Global Structure-from-Motion, C. Tang, O. Wang, P. Tan, In 3DV,2017
- Global Structure-from-Motion by Similarity Averaging, Z. Cui, P. Tan, In ICCV, 2015
Get two sample sequences from Google Drive .
git clone https://github.com/frobelbest/GSLAM.git
Install from http://www.theia-sfm.org
Install from http://ceres-solver.org
Install from https://projects.coin-or.org/Clp
Install from https://opencv.org
Install from https://github.com/stevenlovegrove/Pangolin
Currently, only the xcode project is supplied. You can write your own code to compile on other platforms or wait for future update.
Run on a dataset from Google Drive using GSLAM [sequence_path] [vocabulary_path], for example GSLAM ./robot ./Vocabulary/ORBvoc.txt The ORB vocabulary for loop detection can be downloaded at https://github.com/raulmur/ORB_SLAM2/tree/master/Vocabulary
Under each sequnce folder you will see the following files:
shake.mov
The input video.framestamp.txt
The timestamps for each frame.gyro.txt
The gyroscope readings recorded along with the video.config.yaml
The config settings.
Except the method proposed in the paper, this project also featured in a highly optimized KLT Tracker that can track more than 4000 points on a 1080p video in real-time.
The main bottleneck for this project is the feature tracking, which can be further improved by the paper "Better feature tracking through subspace constraints".
GSLAM was developed at the Simon Fraser University and Adobe. The open-source version is licensed under the GNU General Public License Version 3 (GPLv3). For commercial purposes, please contact cta73@sfu.ca or pingtan@sfu.ca