heyijia / svo_edgelet Goto Github PK

View Code? Open in Web Editor NEW

431.0 21.0 174.0 2.93 MB

A more robust SVO with edgelet feature

License: GNU General Public License v2.0

CMake 4.10% C++ 95.90%

svo_edgelet's Introduction

SVO_edgelet

ChangeList， compare with the origin svo:

add edgelete feature
inition VO: auto select H model and E model
keyframe selection strategy
add pangolin to draw trajectory and remove ROS

install（ubuntu）

Prerequisites

boost - c++ Librairies (thread and system are needed)

sudo apt-get install libboost-all-dev
Eigen 3 - Linear algebra

apt-get install libeigen3-dev
OpenCV3
Sophus - Lie groups we have intergrated sophus in our code. Thanks for the code: https://github.com/strasdat/Sophus.git
g2o - General Graph Optimization OPTIONAL Only required if you want to run bundle adjustment. It is not necessary for visual odometry. I suggest an out-of-source build of g2o:

cd workspace

git clone https://github.com/RainerKuemmerle/g2o.git

cd g2o

mkdir build

cd build

cmake ..

make

sudo make install

Build and Compile

If you want use g2o, please set flag in cmakerlists.txt：

SET(HAVE_G2O TRUE) #TRUE FALSE

compile：

mkdir build cd build cmake .. make

Run code

test file, test_pipline and test_live_vo:

Offline datasets：

To run on tum dataset：test_pipline
To run on euroc dataset：test_piplel_euroc

Online camera：test_live_vo

Results

Authors

Yijia He, Institute of Automation, Chinese Academy of Sciences

Wei Wei, Institute of Computing Technology, Chinese Academy of Sciences

Yanan Gao, Institute of Computing Technology, Chinese Academy of Sciences

svo_edgelet's People

Contributors

Stargazers

Watchers

Forkers

dywuha skylook labimage stephanie-ustc buanderie nemo110110 ayumizll golaced peimengxin hkkoo1 ajaycharan guofengw doctorwk007 wangcoco wangfy1 hitlumino sg47 vcmman xiaoyaoe chenming5828 jiancui1992 maolb mylxiaoyi kongan guwu issac8huxley cumtchenchang qxqxzzz vbillys mocap-tools copark86 ahmadyan yoshua-msc-thesis you2996 iceyu maazmb hackadong wangjio glenbian eagle-feng xunshanman ewenwan simonsroad jamesxy7 zheng618 easonweng li9616 wlt027 cuijianzhu wanghuayou1028 frankszn youruncleda yisenyu irobotdavinci a-l-a-n einsteinliu zlingbryan petertdd shunliuniliu songjiaxina tjevgerres xiehongle firskey kevinhikali feifanrensheng kinggreat24 forrest-z caoyifeng001 jx0913 tangyinzhi eglrp stevencui jxqj linzzmichael dada1118 david-willo slamcn lambdald qingsimon nnu-gisa greatqz zhangnage123 newbeecode zhangxiaoya xuehait bestzed1 why-freedom nuaaguo ynfei hchrobo leatherwang elegracer andylei77 laidongd c-prufrock quitino superzhiwei xlong0513 sagacious-cao haosen9527

svo_edgelet's Issues

小边原理

用的是小边吗？小边用的哪个啊，有什么讲小边原理的教材可以看吗？还有为什么对着白墙可能检测出来呢，麻烦前辈指教

Fail when using customized camera parameter

When I change the parameters at cam_ = new svo::PinholeCamera intest_live_vo.cpp, the VO pipeline always fails with error Not enough matched features
Not enough matched features, Reproject process failure!
The cam param is calculated by Matlab toolbox.
And using the default cam param works just fine.

特征点少并且速度很慢

贺博好。我用了一个无人机航拍数据集，分辨率1920*1080，速度很慢并且特征点奇少。
请问速度慢是图像分辨率的问题么？特征点奇少并且跟丢是什么原因？如下图

Compile problem

Which version is the OpenCV???

fisheye camera model

贺博你好,请问添加鱼眼相机模型除了在camera_model.cpp中参照pinhole模型增加fisheye模型外,还需要修改哪部分的代码吗?关于svo_edgelet添加鱼眼模型,有什么建议吗?十分感谢

Read rectification parameters

请问这几个参数是什么，MH_05 数据集里好像没有这个文件。

Hello, could this code run in windows?? Looking forwards to your reply, thanks!

what is the new of keyframe selection strategy for this method?

Do any body know what the different between this method and original svo in keyframe selection?

KITTI数据集出现段错误？定位错误在在fast检测里，faster_corner_detect_10<false>(img, img_width, img_height, img_stride, barrier, corners);这个函数错误，请问是怎么回事？

matcher.cpp文件里的depthFromTriangulation是否有bug？

代码：Matrix<double,3,2> A; A << T_search_ref.rotation_matrix() * f_ref, f_cur;
这里A矩阵不应该是 -f_cur 吗？是Bug还是我想错了呢？
求指导～～

always fail on kitti dataset and tum dataset

您这里提供的代码在Kitti数据集和TUM 数据集里面基本上都是初始化之后过几帧就挂掉了,很快就跟踪的点数目不够了,请问是我哪里配置不对吗?我测试的是KITTI里面第一个序列还有TUM中fr1_room那个序列.不过fr1_xyz那个序列是可以成功跟踪的.

what is version of pangolin ?

sudo make install pangolin error,
下载/Pangolin-0.4/build$ sudo make install
[sudo] password for ouweiqi:
Inconsistency detected by ld.so: dl-version.c: 224: _dl_check_map_versions: Assertion `needed != ((void *)0)' failed!
make: *** [cmake_check_build_system] 错误 127

appreciate you help me .

Is this equivalent to SVO2.0 in terms of forward facing camera performance?

As it is mentioned in the changelog that new KeyFrame selection criteria as well as automatic selection of E and H matrix is added. Are these chances equivalent to that in SVO2.0 which allows SVO to work on forward facing cameras as well ?
Looking forward to an answer. Thank You!

深度滤波器中更新小边特征方向

if(it->ftr->type == Feature::EDGELET)
{
  it->ftr->grad_cur_ = (matcher_.A_cur_ref_ * it->ftr->grad).normalized();
}

........
........
if(it->ftr->type == Feature::EDGELET)
{
it->ftr->grad = it->ftr->grad_cur_; // edgelete in newkeyframe, it's direction should change;
}

在我的理解中，it->ftr应该是一直储存这个特征被第一次检测到的feature。

深度滤波器这边极线搜索匹配到的edgelet的ftr变量，没有看到被关联到newkeyframe的代码。

不知道是否是我理解有误，还请赐教一下^^

运行test_live_vo时画面卡在第一帧

运行test_live_vo时画面卡在第一帧，且显示Frame-Id：0 #Features：0，请问大家知道这是什么问题吗？

数据集

贺神好，请问有提供数据集下载地址吗？tum里没找到pipline, Euroc 也是。可否告知下数据集下载地址呢？谢谢您哈！

[WARN] Init WARNING: 40 inliers minimum required.

特征太少了,直接挂了
[WARN] Init WARNING: 40 inliers minimum required.

test_live_vo error

hi @HeYijia @supersheepcy
when I run the test_live_vo
there will occur a error like:HIGHGUI ERROR:libv4l unable to ioctl S_FMT
I use the inter aero RTF drone as the platform.
the drone has the intel realsense R200 and a facedown camera on the bottom of the drone.
I want to test this program on both of the camera.
the opencv version is :2.4.13.6
Do anyone have method for this?

Do you have the nonros SVO2.0 version?

This is a very cool repo. I am wondering if you have the nonros version for the newest svo.
Thank you

是单目直接法吗？

我测试的euroc的数据集，但是读取数据时同时读取了左右两个相机的数据？

Problem With Euroc Dataset

Sorry to distrubing you. I have also test it on the euroc dataset MAV_EASY_01, but there still show not enough matched featrues. Reproject process failure! Is there something wrong with the parameters?

svo_edgetlet 依赖哪个版本的opencv?

svo::findEssentialMat(x1, x2, focal, pp, CV_RANSAC, 0.99, 2.0/focal_length, cv::noArray() );

sorry to disturb you.
I want to modify the svo code(on ros) with your code.
but I have a problem with the “svo::findEssentialMat” in five-point.cpp

my kinetic use opencv in version of 3.3.1
when I compile the code it will occur an error in"cv::Mat(tempMask).setTo(true);"
it said:"no matching function for call to 'cv::mat::mat(cvmat*&)'"
Dose this error caused by the version of opencv?
Could you please tell me if I can modified it by using other method?
Thanks for your sharing and patience.

evaluate it with Evo

Hello, which variable stores the position and camera direction of each moment? Would you like to output it and evaluate it with Evo

Can it deal with rolling shutter camera？

where edges feature were used?

Thanks for your open source code. I find that the edge features only exists in processFirstFrame function.
Does it means that only initialization use edge features?
Thanks very much.

轨迹信息

可以获取它的轨迹信息吗，想与真实轨迹信息做一下对比实验？

HIGHGUI ERROR:V4L2:

hi,你好。我运行test_live_vo时，终端显示HIGHGUI ERROR:V4L2:Pixel format of incoming image is unsupported by OpenCV BenchmarkNode finished.我使用的是2.4.13版本的Opencv，请问这是什么问题呢？谢谢

请问怎么运行？

实在是门外汉。课题里面又用到了视觉里程计。想问一下。
编译生成可执行文件后，euroc飞行数据也下了（mav0），运行命令是什么呢？

比如
./test_pipel_euroc , 后边儿应该怎样加上要运行得数据呢？

some errors

Hi,
There are some errors when I ran it,I can't find solutions of this case.Can you tell me what's wrong with it ?Thank you!
What's your opencv version?

opencv支持

请问这个程序只能调用opencv2吗？我用opencv3.4.0编译时会报错

Run mobile phone dataset

Hi
Sorry to disturb you.
Can I run my own dataset on this project?

Thanks

Not enough matched featrues. Reproject process failure!

我用tum的rgbd_dataset_freiburg3_walking_xyz数据集，从头到尾就没有成功过。。。一直显示特征点太少。

编译成功加载测试数据集后trajectory view无反应

您好，打扰了。想请问一下，我这边编译运行成功了，但是下载测试数据集运行test_pipeline后SVO:trajectory viewer中没有任何结果显示可能是什么原因呢？之前在ros上也跑过svo代码也是无法成功加载数据集跑测试，编译之类的都没有报错100%编译通过了。但按照官网加载数据集后roslunch会自动shutdown...

检测梯度最大点

贺博士打扰了，
我在自己的摄像头下（对着特征较少的环境，比如墙面理论上来说这种情况下特征点提取和edgelet提取应该十分有限）跑您的代码过程中并没有发现有检测到梯度变化最大点。
不知道是否是没有做可视化所以看不到，还是因为这部分特征始终没有参与计算。
可否麻烦您稍微讲解一下。
谢谢🙏

error: no matching function for call to ‘format(cv::Mat&, const char [4])’

/home/ubuntu/svo_edgelet-master/src/feature_detection.cpp: In function ‘void svo::feature_detection::saveMatToCsv(cv::Mat, std::string)’:
/home/ubuntu/svo_edgelet-master/src/feature_detection.cpp:28:39: error: no matching function for call to ‘format(cv::Mat&, const char [4])’
28 | outputFile <<cv::format(data,"CSV")<<endl;
| ^
In file included from /usr/local/include/opencv4/opencv2/core.hpp:3302,
from /usr/local/include/opencv4/opencv2/opencv.hpp:52,
from /home/ubuntu/svo_edgelet-master/include/svo/global.h:28,
from /home/ubuntu/svo_edgelet-master/include/svo/feature_detection.h:20,
from /home/ubuntu/svo_edgelet-master/src/feature_detection.cpp:17:
/usr/local/include/opencv4/opencv2/core/operations.hpp:430:19: note: candidate: ‘cv::String cv::format(const char*, ...)’
430 | CV_EXPORTS String format( const char* fmt, ... ) CV_FORMAT_PRINTF(1, 2);
| ^~~~~~
/usr/local/include/opencv4/opencv2/core/operations.hpp:430:39: note: no known conversion for argument 1 from ‘cv::Mat’ to ‘const char*’
430 | CV_EXPORTS String format( const char* fmt, ... ) CV_FORMAT_PRINTF(1, 2);
| ~~~~~~~~~~~~^~~
/usr/local/include/opencv4/opencv2/core/operations.hpp:435:16: note: candidate: ‘cv::Ptrcv::Formatted cv::format(cv::InputArray, cv::Formatter::FormatType)’
435 | Ptr format(InputArray mtx, Formatter::FormatType fmt)
| ^~~~~~
/usr/local/include/opencv4/opencv2/core/operations.hpp:435:61: note: no known conversion for argument 2 from ‘const char [4]’ to ‘cv::Formatter::FormatType’
435 | Ptr format(InputArray mtx, Formatter::FormatType fmt)
| ~~~~~~~~~~~~~~~~~~~~~~^~~
/home/ubuntu/svo_edgelet-master/src/feature_detection.cpp: In member function ‘virtual void svo::feature_detection::EdgeDetector::detect(svo::Frame*, const ImgPyr&, double, svo::Features&)’:
/home/ubuntu/svo_edgelet-master/src/feature_detection.cpp:545:15: warning: comparison of integer expressions of different signedness: ‘int’ and ‘std::vectorsvo::feature_detection::Edgelete::size_type’ {aka ‘long unsigned int’} [-Wsign-compare]
545 | int a = (n>edge_list.size())? edge_list.size():n;
| ~^~~~~~~~~~~~~~~~~
make[2]: *** [CMakeFiles/svo.dir/build.make:303：CMakeFiles/svo.dir/src/feature_detection.cpp.o] 错误 1
make[1]: *** [CMakeFiles/Makefile2:179：CMakeFiles/svo.dir/all] 错误 2
make: *** [Makefile:84：all] 错误 2

build demo with undefined reference to Sophus

[ 82%] Built target svo
[ 85%] Linking CXX executable ../bin/test_pipeline
../lib/libsvo.so：对‘Sophus::SE3::operator*(Eigen::Matrix<double, 3, 1, 0, 3, 1> const&) const’未定义的引用
../lib/libsvo.so：对‘Sophus::SO3::log() const’未定义的引用
../lib/libsvo.so：对‘Sophus::SE3::operator=(Sophus::SE3 const&)’未定义的引用
../lib/libsvo.so：对‘Sophus::SE3::SE3()’未定义的引用
../lib/libsvo.so：对‘Sophus::SE3::SE3(Eigen::Matrix<double, 3, 3, 0, 3, 3> const&, Eigen::Matrix<double, 3, 1, 0, 3, 1> const&)’未定义的引用
../lib/libsvo.so：对‘Sophus::SE3::exp(Eigen::Matrix<double, 6, 1, 0, 6, 1> const&)’未定义的引用
../lib/libsvo.so：对‘Sophus::SE3::SE3(Sophus::SE3 const&)’未定义的引用
../lib/libsvo.so：对‘Sophus::SE3::inverse() const’未定义的引用
../lib/libsvo.so：对‘Sophus::SE3::operator*(Sophus::SE3 const&) const’未定义的引用
../lib/libsvo.so：对‘Sophus::SO3::matrix() const’未定义的引用
collect2: error: ld returned 1 exit status
CMakeFiles/test_pipeline.dir/build.make:141: recipe for target '../bin/test_pipeline' failed
make[2]: *** [../bin/test_pipeline] Error 1
CMakeFiles/Makefile2:104: recipe for target 'CMakeFiles/test_pipeline.dir/all' failed
make[1]: *** [CMakeFiles/test_pipeline.dir/all] Error 2
Makefile:83: recipe for target 'all' failed
make: *** [all] Error 2

I have installed Sophus already, what could be the probblem? Thanks in advance.