Schur_SP_VINS

This project introduces modifications to the visual frontend of SchurVINS in order to enhance the performance of feature extraction and tracking.

Reproduction of SchurVINS

Here are some tips from my experience reproducing SchurVINS. Additionally, if the download speed of the SchurVINS repository is too slow, you may consider switching to a different mirror.

The project was run in an environment consisting of Ubuntu 20.04, ROS Noetic, and OpenCV 3.4.20 (Ubuntu 18.04 + ROS Melodic + OpenCV 3.2.0 has also been tested).

First, you need to open three terminals and run the following commands in each of them respectively.

cd /home/<your_usr_name>/SchurVINS-ws
source devel/setup.bash
roslaunch svo_ros euroc_vio_stereo.launch

cd /home/<your_usr_name>/SchurVINS-ws
source devel/setup.bash
cd /home/<your_usr_name>/SchurVINS-ws/src/SchurVINS
rviz -d svo_ros/rviz_config.rviz

cd /home/<your_usr_name>/datasets/rosbag
rosbag play MH_01_easy.bag

Run and Evaluation

In this project, the traditional hand-crafted feature extraction approach was replaced by the deep learning-based SuperPoint method. However, no changes were made to the execution process.

Dataset Processing

The publicly available EuRoC dataset is used in this project and needs to be converted into the following format:

<dataset_name>
├── calib.yaml  --> 校准文件
├── data
│   ├── groundtruth_matches.txt
│   ├── groundtruth.txt
│   ├── stamped_groundtruth.txt
│   ├── images.txt
│   ├── img --> 包含实际图像的文件夹
│   ├── imu.txt
└── dataset.yaml --> 可选，用于指定特定数据集的参数

The dataset has already been processed and is ready for use. It is located in the svo_benchmarking folder.

Execution Configuration

Open two terminals, one to run:

roscore

and the other to run:

rosrun svo_benchmarking benchmark.py superpoint.yaml

There is no need to launch a separate terminal for roslaunch, as benchmark.py already starts the necessary ROS nodes.

Evaluation

rosrun rpg_trajectory_evaluation analyze_trajectory_single.py <result_folder>

For each evaluation, make sure to copy the labeled ground truth file into the results directory.