IS-VINS is a sparisification-based system which extends VINS-Mono to consistent optimization of VO & VIO & Pose-Graph; This system is motivated by preserving sparse and nonlinear information after marginalization and simplifying VIO scheme rather than preserving linearied prior that ordinary fixed-lag smoother shares;Thanks to sparisification,pose graph is builted upon information structure to achieve consistent optimazation;This code supports Linux without ROS, it is possible to run on Mac OS X or Windows with sightly change.Please cite us if you use our code.
Followings are main contributions in this work:
-
Frontend is optic flow based method just like VINS-MONO but simplified.
-
VIO is performed in two stages marginalization and sparsification,which turns VIO to be a combanation of VO and VIO.
-
Pose graph optimization reuses information of VO with gravity observation and relative pose prior and performs with loop-closure information. So it's able to evaluate covariance of camera pose online.
Authors：: Jixiang Ma(unicorn@hust.edu.cn)
We tested this code with ubuntu 18.04 with Ceres 2.0.0,Eigen 3.3.4,OpenCV 3.2.0 and Pangolin.
$ git clone https://github.com/lyeemax/IS-VINS.git
$ mkdir build
$ cd build/
$ cmake ..
$ make -j 9
Download EuRoc Mav dataset and extract zips. Open a terminal with parameters of dataset path and config files:
$ ./run_euroc PATH_TO_EUROC/mav0/ ../config/