This repository contains the code for our final project for EECS 568: Mobile Robotics: Methods and Algorithms. Our project compared two different SLAM algorithms, ORB-SLAM3 and OKVIS. Then, we examined the effects of real-time image enhancement on both algorithms.
Here is our paper (TODO).
Here is our final presentation and video.
There are three separate projects within this repository: ORB-SLAM3 with Image Enhancement, OKVIS with Image Enhancement, and Single Image Enhancement. We used the Aqualoc dataset to test all of our projects. In particular, we made use of Harbor_01 sequence. To run the example dataset, it needs to be in the correct format, which can be downloaded here. You can also download it from source and manually convert it.
ORB_SLAM3_modified, okvis and image_enhancement folders are for the three projects. util folder contains files for checking matched features, extracting timestamps, renaming images from frame number to timestamp, and changing frame number in groundtruth file to timestamp. data folder contains some results we obtained. evaluation folder contains files for evaluation.
The first project is ORB_SLAM3 modified to include image enhancement capabilities. To build, first install dependencies by following instructions from ORB_SLAM3.
- Pangolin
- OpenCV 4.2.0
- OpenCV Contrib
- Eigen3
Then, to build our ORB_SLAM3_modified project:
cd ORB_SLAM3_modified
chmod +x build.sh
./build.sh
To run the project. For example, on harbor dataset:
./Examples/Monocular-Inertial/mono_inertial_harbor ./Vocabulary/ORBvoc.txt <path_to_config_file> <path_to_dataset_folder> <path_to_timestamps> <output_filename> [image_method]
Here, the [image_method] is an optional image enhancement parameter which is set to 'none' by default. To enable image enhancement, set this parameter to 'simple'. The same convention is also followed in the case of OKVIS.
The second project is OKVIS modified to include image enhancement capabilities. To build, first install dependencies:
- CMake:
sudo apt-get install cmake - Google-Glog and GFlags:
sudo apt-get install libgoogle-glog-dev - BLAS and LAPACK:
sudo apt-get install libatlas-base-dev - Eigen3:
sudo apt-get install libeigen3-dev - SuiteSparse and CXSparse:
sudo apt-get install libsuitesparse-dev - Boost:
sudo apt-get install libboost-dev libboost-filesystem-dev - OpenCV 4.2.0:
sudo apt-get install libopencv-dev
Then, from the root directory of the repository:
- Go into the OKVIS project:
cd okvis - Create the build directory:
mkdir build - Generate the Makefile:
cmake .. - Build the project:
make -j
The project can now be run from the build directory as:
./okvis_app_synchronous <path_to_okvis_config_yaml> <path_to_dataset> <image_enhancement_technique>
Then, run the file on the example harbor dataset as:
./okvis_app_synchronous ../config/config_harbor.yaml <path_to_transformed_harbor_dataset>
Last is the image enhancement project, which is a simple project that can be used to test various image enhancement algorithms on single images at a time, to see if features and contrast will be improved. First, install dependencies:
- CMake:
sudo apt-get install cmake - OpenCV 4.2.0:
sudo apt-get install libopencv-dev
Then, from the root directory of this repository:
- Go into the Image Enhancement project:
cd image_enhancement - Create the build directory:
mkdir build && cd build - Generate the Makefile:
cmake .. - Build the project:
make -j
The project can be run from the build directory as ./image_enhancement <path_to_input_image> <path_to_output_image>
To re-implement the evaluation of trajectory estimation of ORB-SLAM3 or OKVIS, run the jupyter notebook at Mobile_Robotics_Team22/evaluation/Demo_Evaluation_Result.ipynb.