chauchorhim / fast-svo

fast-SVO

fast-SVO is a basic stereo visual odometry. A python version SVO implemented during the course 02504 (Computer Vision F21) in DTU and a popular SLAM structure ORB-SLAM give inspiration for the SVO design and C++ implementation style for this project. Feature detection and matching parts use inline functions provided by OpenCV. P3P solution provided by Kneip http://rpg.ifi.uzh.ch/docs/CVPR11_kneip.pdfis is implemented in this project.

Prerequisites

This project is implemented and teseted in Ubuntu 18.04.

C++11 Compiler

This project use std::thread, std::chrono.

OpenCV

This project uses the latest OpenCV 4.5.1.

Eigen3

This project uses version 3.3.91.

Example

KITTI Dataset

Download the dataset (grayscale images) from http://www.cvlibs.net/datasets/kitti/eval_odometry.php
The setting files are stored in ${PROJECT_SOURCE_DIR}/Example/KITTI setting/
Usage example:

./Example/Stereo/stereo_kitti Example/KITTI_setting/KITTI04-12.yaml PATH_TO_DATASET_FOLDER/dataset/sequences/ 07 

Evaluation

Evaluation comparison between this project and my previous python version on KIITI dataset (grayscale 07)