VIO

This code implements a semi-direct monocular visual odometry pipeline which has been integrated with a EKF to infused IMU and CMD as well. The feature extraction and image alignment part are implemented in OpenCL to take advantage of GPU.

Documentation

OpenCL documentation

Instructions

Please be sure that you have the OpenCL driver installed

Setting

You can edit the algorithm parameters in the follwoing file, please look at the config file to underestand the meaning of the parameters (CPU_version/svo/include/svo/config.h)

CPU_version/vio/param/vo_fast.yaml

the GPU_version

The dependencies are:

OpenCV 4
Eigen
G2o # build g2o with -DG2O_HAVE_OPENGL=ON -DBUILD_WITH_MARCH_NATIVE=ON # G2o version => https://github.com/RainerKuemmerle/g2o/tree/memory_management
Boost
OpenCL

known issues

1. The estimated oriantetion is not accurate 
2. There is a scale map issue in the algorithm 
3. The number of matched points are increasing a lot we need to limit them in a way that will not cause some error in the estimated odometry