gaoxiang12 / orb-ygz-slam Goto Github PK

高博,您好! ORB-SLAM2有一个视觉词典模块, 是用于将ORB特征的词袋向量比对的. 但是如果换成fast特征点,那这个视觉词典模块就没有作用了?

When i run with the TUM data sets, it can start but mostly it will stop and the terminal do nothing , and I should input ctrl+c to stop the terminal. And in a few cases, it will complete running and the terminal will show a lot of "Try relocalization" and "Set state to Lost",and finally show" Illegal instruction (core dumped)". I want to know why did this happen and what can I do to solve it?
I have been trying to solve this, but i can't. And this may be useful.
ulimit -a
core file size (blocks, -c) unlimited
data seg size (kbytes, -d) unlimited
scheduling priority (-e) unlimited
file size (blocks, -f) unlimited
pending signals (-i) unlimited
max locked memory (kbytes, -l) unlimited
max memory size (kbytes, -m) unlimited
open files (-n) 1024
pipe size (512 bytes, -p) 8
POSIX message queues (bytes, -q) unlimited
real-time priority (-r) unlimited
stack size (kbytes, -s) unlimited
cpu time (seconds, -t) unlimited
max user processes (-u) unlimited
virtual memory (kbytes, -v) unlimited
file locks (-x) unlimited
but it still couldn't work.
Thank you very much.

您好！请问sophus的编译 是直接将下载的sophus安装在/ORB-YGZ-SLAM/Thirdparty/sophus下面吗？
如果您看到，望回复，十分感谢！！！

Nice work! I wanted to try on the code but got error when compiling. Installation of project by running ./generate.sh.

System: Ubuntu 17.04
OpenCV: 3 from github master
dependency all installed

The error as follows:

[ 65%] Built target ygz-SLAM
[ 68%] Linking CXX executable ../Examples/Monocular/mono_tum
[ 70%] Linking CXX executable ../Examples/Monocular/mono_euroc
[ 75%] Linking CXX executable ../Examples/RGB-D/rgbd_tum
[ 75%] Linking CXX executable ../Examples/Monocular/mono_kitti
../lib/libygz-SLAM.so: undefined reference to cv::read(cv::FileNode const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)' collect2: error: ld returned 1 exit status CMakeFiles/rgbd_tum.dir/build.make:153: recipe for target '../Examples/RGB-D/rgbd_tum' failed make[2]: *** [../Examples/RGB-D/rgbd_tum] Error 1 ../lib/libygz-SLAM.so: undefined reference to cv::read(cv::FileNode const&, std::__cxx11::basic_string<char, std::char_traits, std::allocator >&, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&)'
collect2: error: ld returned 1 exit status
CMakeFiles/Makefile2:215: recipe for target 'CMakeFiles/rgbd_tum.dir/all' failed
make[1]: *** [CMakeFiles/rgbd_tum.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs....
../lib/libygz-SLAM.so: undefined reference to cv::read(cv::FileNode const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)' collect2: error: ld returned 1 exit status CMakeFiles/mono_euroc.dir/build.make:153: recipe for target '../Examples/Monocular/mono_euroc' failed make[2]: *** [../Examples/Monocular/mono_euroc] Error 1 CMakeFiles/mono_tum.dir/build.make:153: recipe for target '../Examples/Monocular/mono_tum' failed make[2]: *** [../Examples/Monocular/mono_tum] Error 1 CMakeFiles/Makefile2:67: recipe for target 'CMakeFiles/mono_euroc.dir/all' failed make[1]: *** [CMakeFiles/mono_euroc.dir/all] Error 2 CMakeFiles/Makefile2:141: recipe for target 'CMakeFiles/mono_tum.dir/all' failed make[1]: *** [CMakeFiles/mono_tum.dir/all] Error 2 ../lib/libygz-SLAM.so: undefined reference to cv::read(cv::FileNode const&, std::__cxx11::basic_string<char, std::char_traits, std::allocator >&, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&)'
collect2: error: ld returned 1 exit status
CMakeFiles/mono_kitti.dir/build.make:153: recipe for target '../Examples/Monocular/mono_kitti' failed
make[2]: *** [../Examples/Monocular/mono_kitti] Error 1
CMakeFiles/Makefile2:104: recipe for target 'CMakeFiles/mono_kitti.dir/all' failed
make[1]: *** [CMakeFiles/mono_kitti.dir/all] Error 2
Makefile:83: recipe for target 'all' failed
make: *** [all] Error 2

Thank you so much!

精度方面，单目的效果和okvis/VINS之间有没有进行对比过？

Hi Everyone, I was wondering if someone has a data collection app which can be used to dump imu/images from a ,mobile phone and that data has been tested with YGZ-SLAM. And if so, will you be able to share the code.

I have tried with Samsung S8 and did the calibration using Kalibr but yet to see success with that data.

Thanks!

这修改成直接法，能用卷帘快门式的相机(C920)吗?

test_LK_tracking.cpp:(.text.startup+0x6f8)：对‘Sophus::SE3::SE3()’未定义的引用
test_LK_tracking.cpp:(.text.startup+0xee9)：对‘Sophus::SE3::SE3()’未定义的引用
../../lib/libygz-algorithm.so：对‘Sophus::SO3::exp(Eigen::Matrix<double, 3, 1, 0, 3, 1> const&)’未定义的引用
../../lib/libygz-algorithm.so：对‘Sophus::SE3::exp(Eigen::Matrix<double, 6, 1, 0, 6, 1> const&)’未定义的引用
../../lib/libygz-algorithm.so：对‘Sophus::SO3::log() const’未定义的引用
../../lib/libygz-algorithm.so：对‘Sophus::SE3::SE3(Sophus::SO3 const&, Eigen::Matrix<double, 3, 1, 0, 3, 1> const&)’未定义的引用
../../lib/libygz-algorithm.so：对‘Sophus::SE3::operator=(Sophus::SE3 const&)’未定义的引用
../../lib/libygz-algorithm.so：对‘Sophus::SE3::SE3(Sophus::SE3 const&)’未定义的引用
../../lib/libygz-algorithm.so：对‘Sophus::SE3::inverse() const’未定义的引用
../../lib/libygz-algorithm.so：对‘Sophus::SE3::log() const’未定义的引用
../../lib/libygz-algorithm.so：对‘Sophus::SE3::operator*(Sophus::SE3 const&) const’未定义的引用
../../lib/libygz-algorithm.so：对‘Sophus::SO3::matrix() const’未定义的引用
../../lib/libygz-algorithm.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/libygz-algorithm.so：对‘Sophus::SE3::matrix() const’未定义的引用
../../lib/libygz-algorithm.so：对‘Sophus::SE3::operator*(Eigen::Matrix<double, 3, 1, 0, 3, 1> const&) const’未定义的引用
collect2: error: ld returned 1 exit status
test/CMakeFiles/test_LK_tracking.dir/build.make:182: recipe for target '../bin/test_LK_tracking' failed
make[2]: *** [../bin/test_LK_tracking] Error 1
CMakeFiles/Makefile2:238: recipe for target 'test/CMakeFiles/test_LK_tracking.dir/all' failed
make[1]: *** [test/CMakeFiles/test_LK_tracking.dir/all] Error 2
Makefile:83: recipe for target 'all' failed
make: *** [all] Error 2

R.T
Hi Xiang, as in official ORB-SLAM2, there's ORS package for us to build a ros wrapper for my own camera usage. Could you please tell me how to create the same thing according to your realization? @gaoxiang12 Thank you so much!

Hi guys,
Thank you for sharing your code! But maybe you put a wrong config file in the repo....(no IMU params). Waiting for the update.

Regards

in localmapping.cc
TryInitVIO:
fgw.open("/home/jp/opensourcecode/ORB_SLAM2/tmp/gw.txt");
fscale.open("/home/jp/opensourcecode/ORB_SLAM2/tmp/scale.txt");
fbiasa.open("/home/jp/opensourcecode/ORB_SLAM2/tmp/biasa.txt");
fbiasg.open("/home/jp/opensourcecode/ORB_SLAM2/tmp/biasg.txt");

how can I get gw.txt scale.txt biasa.txt?

I0423 10:45:20.048502 19680 Tracking.cc:1638] Search by projection returns 53
Stack trace (most recent call last):
#8 Object "", at 0xffffffffffffffff, in
#7 Object "/home/wh/Desktop/mygitlab/svo-orb-1/Examples/Stereo/stereo_euroc_line", at 0x5629369fb3bd, in _start
#6 Source "../csu/libc-start.c", line 308, in __libc_start_main [0x7fc53d73c082]
#5 Source "/home/wh/Desktop/mygitlab/svo-orb-1/Examples/Stereo/stereo_euroc_line.cc", line 180, in main [0x5629369fc626]
177: #endif
178:
179: // Pass the images to the SLAM system
> 180: SLAM.TrackStereo(imLeftRect,imRightRect,tframe);
181:
182: #ifdef COMPILEDWITHC11
183: std::chrono::steady_clock::time_point t2 = std::chrono::steady_clock::now();
#4 Source "/home/wh/Desktop/mygitlab/svo-orb-1/src/System.cc", line 242, in TrackStereo [0x7fc5420daa1d]
239: }
240: }
241: //用矩阵Tcw来保存估计的相机 位姿，运动追踪器的GrabImageStereo函数才是真正进行运动估计的函数
> 242: SE3f Tcw = mpTracker->GrabImageStereo(imLeft, imRight, timestamp);
243: //给运动追踪状态上锁
244: unique_lock lock2(mMutexState);
245: //获取运动追踪状态
#3 Source "/home/wh/Desktop/mygitlab/svo-orb-1/src/Tracking.cc", line 263, in GrabImageStereo [0x7fc542105a72]
260: mCurrentFrame = Frame(mImGray, imGrayRight, timestamp, mpORBextractorLeft, mpORBextractorRight, mpORBVocabulary,
261: mK, mDistCoef, mbf, mThDepth);
262: // Step 3 ：跟踪
> 263: Track();
264: //返回位姿
265: return mCurrentFrame.mTcw;
266: }
#2 Source "/home/wh/Desktop/mygitlab/svo-orb-1/src/Tracking.cc", line 525, in Track [0x7fc54210703d]
522: bOK = TrackLocalMap();
523: } else {
524: // 未提特征，用Direct方法比较
> 525: bOK = TrackLocalMapDirect();
526: if (!bOK) {
527: // Track local map 数量少，可能是前一步位姿估计的不准，也可能是视觉上匹配的就是太少
528: // clear the features and try search in local map
#1 Source "/home/wh/Desktop/mygitlab/svo-orb-1/src/Tracking.cc", line 2180, in TrackLocalMapDirect [0x7fc54210f56f]
2177: if (mSensor == System::RGBD)
2178: mCurrentFrame.ComputeStereoFromRGBD(mCurrentFrame.mImDepth);
2179: else if (mSensor == System::STEREO)
>2180: mCurrentFrame.ComputeStereoMatches();
2181:
2182: // 优化当前帧的位姿，确定 inlier
2183: // Optimize pose
#0 Source "/home/wh/Desktop/mygitlab/svo-orb-1/src/Frame.cc", line 671, in ComputeStereoMatches [0x7fc5424709ee]
668: }
669:
670: sort(vDistIdx.begin(), vDistIdx.end());
> 671: const float median = vDistIdx[vDistIdx.size() / 2].first;
672: const float thDist = 1.5f * 1.4f * median;
673:
674: for (int i = vDistIdx.size() - 1; i >= 0; i--) {
段错误 (地址没有映射到对象 [(nil)])

Process finished with exit code 139 (interrupted by signal 11: SIGSEGV)

When i run with the TUM dataests, it can start but it will immediately exit and the terminal show " Illegal instruction (core dumped)" .I want to know why did this happen and what can i do to solve it?
Thank you very much.

Hello,

My name is Yoni. We're researching the topic of fusion of SLAM and IMU.
I encountered your work on ORB-YGZ-SLAM, and I wondered:

1. Is the implementation of SVO tracking meant only to increase speed? Or do you believe that it has more advantages?
2. When you say that you followed Raul's paper when you implemented monocular ORB SLAM + IMU, do you mean this paper 'Visual-Inertial Monocular SLAM with Map Reuse' ?
3. If so, do you maybe have (and are willing to share) an older version of your code, in which monocular ORB SLAM + IMU is already implemented, but with the original ORB SLAM tracking?

Thank you so much.

Thank you for your open source.
I try to turn on the IMU , and run it with EUROC data. But when IMU inited, get the error:

W0716 23:09:22.977699 57503 LocalMapping.cc:280] bgest: -0.0215912 00.0315326 000.086801
W0716 23:09:22.978873 57503 LocalMapping.cc:343] -0.0232417 00-9.34961 00.0498976 00-3.28288, gw:0-9.34961 0.0498976 0-3.28288, |gw|=9.90934
W0716 23:09:22.978936 57503 LocalMapping.cc:416] -0.000233174 02.80843e-07 -1.95786e-06 0000.0925712 00.000853654 00-0.0362108, s:-0.000233174, dthetaxy:02.80843e-07 -1.95786e-06, dbiasa:000.0925712 0.000853654 0-0.0362108
W0716 23:09:22.978967 57503 LocalMapping.cc:425] gw: 0-9.25588 0.0493945 0-3.24998, |gw|=9.81
Time: 1.40372e+09, sstar: -0.0232417, s: -0.000233174
I0716 23:09:22.979120 57503 LocalMapping.cc:455] 1.40372e+09, 1.40372e+09, time elapsed: 0
Local Mapping STOP
I0716 23:09:22.981716 57503 LocalMapping.cc:555] 
... Map scale updated ...

I0716 23:09:22.981734 57503 LocalMapping.cc:627] 
... Map NavState updated ...

I0716 23:09:22.981739 57503 LocalMapping.cc:150] set first vins inited : 1
Local Mapping RELEASE
I0716 23:09:22.984490 57503 LocalMapping.cc:642] finish global BA after vins init
I0716 23:09:22.984508 57503 LocalMapping.cc:647] request local mapping stop
Local Mapping STOP
I0716 23:09:22.986210 57503 LocalMapping.cc:652] local mapping stopped
Map updated!
Local Mapping RELEASE
VINS inited, quit VINS init thread
Quit VINSInitThread
I0716 23:09:23.010887 57502 Tracking.cc:1310] Track local map direct with imu failed, inliers: 0
I0716 23:09:23.010916 57502 Tracking.cc:508] Track Local map direct with IMU failed. VINS first inited: 1, mapupdated:1
I0716 23:09:23.010921 57502 Tracking.cc:531] Set vision weak = true
I0716 23:09:23.011090 57502 Tracking.cc:605] Try using imu predicted pose
I0716 23:09:23.064316 57502 Tracking.cc:2090] Failed. return back to feature methods
I0716 23:09:23.073400 57502 Tracking.cc:1091] Search a wider window
I0716 23:09:23.073449 57502 Tracking.cc:1097] Match is not enough, failed.
I0716 23:09:23.074784 57502 LocalMapping.cc:150] set first vins inited : 0
Segmentation fault (core dumped)

hoping for your answer?when turn off the IMU, everything goes well

Hi，Gao Bo
There is no makefile in sophus. And g2o cannot compile successfully.

@gaoxiang12
Hi, I was wondering if this code has been successfully tested with any data sets from Mobile devices specifically android phones? I am trying to use data from S7/S8 but the trajectory flies away right after the IMU pre-integration is done and the map scale is updated.

This is the log I get right after it notifies that scale has been updated.

... Map scale updated ...

I1006 17:02:54.859149 29501 LocalMapping.cc:151] set first vins inited : 1
I1006 17:02:54.872421 29496 Tracking.cc:1481] Track local map direct with imu failed, inliers: 0
I1006 17:02:54.872447 29496 Tracking.cc:604] Track Local map direct with IMU failed. VINS first inited: 1, mapupdated:0
I1006 17:02:54.872462 29496 Tracking.cc:631] Set vision weak = true
I1006 17:02:54.872568 29496 Tracking.cc:728] Try using imu predicted pose
I1006 17:02:54.903020 29496 Tracking.cc:753] In the ELSE condition ... at line 753 ... bOK = 1
I1006 17:02:54.930971 29496 Tracking.cc:1476] Track local map direct with imu failed, inliers: 0
I1006 17:02:54.931025 29496 Tracking.cc:604] Track Local map direct with IMU failed. VINS first inited: 1, mapupdated:0
I1006 17:02:54.931030 29496 Tracking.cc:631] Set vision weak = true
I1006 17:02:54.931133 29496 Tracking.cc:728] Try using imu predicted pose
I1006 17:02:54.959718 29496 Tracking.cc:753] In the ELSE condition ... at line 753 ... bOK = 1
I1006 17:02:54.972455 29496 Tracking.cc:2433] Failed. return back to feature methods
I1006 17:02:54.991106 29496 Tracking.cc:1254] Search a wider window
I1006 17:02:54.991160 29496 Tracking.cc:1260] Match is not enough, failed.

P.S. I used Kalibr for calibrating Camera-IMU

Thanks for your sharing at first.
As you may know, in EuRoC V102 dataset, there are continuous rotational motions in the middle part, which I think could pose great challenges to a mono-based slam system. Meanwhile, when IMU information is fused, the estimation progress should have been robust. So I build your algorithm on my Ubuntu 14.04 machine with configurations as Intel® Core™ i5-3230M CPU @ 2.60GHz × 4, 8G memory. The result surprises me, because the running with IMU module off is better than with IMU on, though there are several failures in pure visual mode, the loop closure function is available and can successfully relocalize. However, when IMU is on, the state just flies out of the vicon room, which I thought would not possibly happen. Below is the runnings.
IMU on:

IMU off:

What are the possible reasons? Look forward to hearing from you.

Thanks for the awesome work! The code is well written with high readability!

Here is a problem for me. Though I built the project successfully, when running the example of EuRoC, I failed.

My config:
Ubuntu: 17.04
OpenCV: 3 from github master (after installing ros seems like the computer starts to use ros-lunar-opencv3, as in error shows below)
ROS: Lunar (but I did not use ROS)

Here is the error:
czm@czm:~/SLAM/ORB-YGZ-SLAM$ ./Examples/Monocular/mono_euroc Vocabulary/ORBvoc.bin Examples/Monocular/EuRoC.yaml /home/czm/Dataset/EuRoC_MAV/mav0/cam0/data Examples/Monocular/EuRoC_TimeStamps/MH01.txt
WARNING: Logging before InitGoogleLogging() is written to STDERR
I0729 10:18:43.053556 3470 System.cc:107] This is YGZ-SLAM, forked from ORB-SLAM proposed by Raul Mur-Artal, University of Zaragoza.
This program comes with ABSOLUTELY NO WARRANTY;
This is free software, and you are welcome to redistribute it
under certain conditions. See ORB-SLAM's LICENSE.txt.

I0729 10:18:43.053719 3470 System.cc:114] Input sensor was set to:
I0729 10:18:43.053740 3470 System.cc:117] Monocular
I0729 10:18:43.100296 3470 System.cc:132]
Loading ORB Vocabulary. This should be super fast ...
I0729 10:18:43.571415 3470 System.cc:148] Vocabulary loaded!

Parameters:
VINS initialize time: 0
Discart time in test data: 0
save tmp file in
open rosbag:
imu topic:
image topic:
local window size: 0
timestamp image delay to imu: 0
Tbc inited:
1 0 0 0
0 -1 0 0
0 0 -1 0
0 0 0 1
[1, 0, 0, 0;
0, -1, 0, 0;
0, 0, -1, 0;
0, 0, 0, 1]
Tcb inited:
1 0 0 -0
0 -1 0 0
0 0 -1 0
0 0 0 1
[1, 0, 0, -0;
0, -1, 0, 0;
0, 0, -1, 0;
0, 0, 0, 1]
TbcTcb:
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
[1, 0, 0, 0;
0, 1, 0, 0;
0, 0, 1, 0;
0, 0, 0, 1]
whether acc9.8? 0/1: 0
whether use IMU 0/1: 0

Camera Parameters:

• fx: 458.654
• fy: 457.296
• cx: 367.215
• cy: 248.375
• k1: -0.283408
• k2: 0.0739591
• p1: 0.00019359
• p2: 1.76187e-05
• fps: 20
• color order: RGB (ignored if grayscale)

ORB Extractor Parameters:

• Number of Features: 1000
• Scale Levels: 4
• Scale Factor: 2
• Initial Fast Threshold: 20
• Minimum Fast Threshold: 7
  I0729 10:18:43.631961 3470 LocalMapping.cc:751] mnLocalWindowSize:0, mbUseIMU:0

Start processing sequence ...
Images in the sequence: 3682

OpenCV Error: Assertion failed (0 <= _rowRange.start && _rowRange.start <= _rowRange.end && _rowRange.end <= m.rows) in Mat, file /tmp/binarydeb/ros-lunar-opencv3-3.2.0/modules/core/src/matrix.cpp, line 483
terminate called after throwing an instance of 'cv::Exception'
what(): /tmp/binarydeb/ros-lunar-opencv3-3.2.0/modules/core/src/matrix.cpp:483: error: (-215) 0 <= _rowRange.start && _rowRange.start <= _rowRange.end && _rowRange.end <= m.rows in function Mat

Aborted (core dumped)

Thanks for the help!

Hi!
I'm tring to build ORB-YGZ-SLAM, but there isn't CMakeLists.txt in Thirdparty/sophus so I cannot build it...

cd ../../sophus
mkdir build
cd build
cmake ..

I'm stuck on this command cmake ..

Does this project only support Intel x86-64 architecture?

`Build type: Release
-- Using flag -std=c++11.
-- Could NOT find CHOLMOD (missing: CHOLMOD_INCLUDE_DIR CHOLMOD_LIBRARIES)
CMake Error: The following variables are used in this project, but they are set to NOTFOUND.
Please set them or make sure they are set and tested correctly in the CMake files:
CHOLMOD_INCLUDE_DIR
used as include directory in directory /home/sunkaikai/skk/zlB/ORB-YGZ-SLAM
used as include directory in directory /home/sunkaikai/skk/zlB/ORB-YGZ-SLAM
used as include directory in directory /home/sunkaikai/skk/zlB/ORB-YGZ-SLAM
used as include directory in directory /home/sunkaikai/skk/zlB/ORB-YGZ-SLAM
used as include directory in directory /home/sunkaikai/skk/zlB/ORB-YGZ-SLAM
used as include directory in directory /home/sunkaikai/skk/zlB/ORB-YGZ-SLAM
used as include directory in directory /home/sunkaikai/skk/zlB/ORB-YGZ-SLAM
used as include directory in directory /home/sunkaikai/skk/zlB/ORB-YGZ-SLAM
CHOLMOD_LIBRARY
linked by target "ygz-SLAM" in directory /home/sunkaikai/skk/zlB/ORB-YGZ-SLAM

-- Configuring incomplete, errors occurred!
`
would you please have a look? There might be some problems with g2o?

As is shown in LoopClosing.cc, Loop Detection is off in IMU mode.
Does any one notice this and understand why?

            // Check if there are keyframes in the queue
            if (CheckNewKeyFrames()) {
                // Detect loop candidates and check covisibility consistency
                if (DetectLoop()) {
                    if (!mbUseIMU) {
                        // Compute similarity transformation [sR|t]
                        // In the stereo/RGBD case s=1
                        if (ComputeSim3()) {
                            // Perform loop fusion and pose graph optimization
                            LOG(INFO) << "Find a loop, correct it" << endl;
                            CorrectLoop();
                        }
                    }
                }
            }

预积分里面的第一部分噪音传播和biases更新看见了，但是IMU　factor: residual errors and Jacobians似乎没有，这是不是一个精简版的预计分？

After installation, I had ONCE successfully run the mono_euroc_vins.
But I couldn't execute it with GUI again. (top shows that the program is running, but I cannot found GUI form.)

Thanks Dr. Gao for sharing the codes!

Ubunut14.04, OpenCV: 3.1.0, Eigen: 3.2.10.

It works well in "mono_euroc", but failure in "stereo_euroc".

Take the V101 for example:

In "stereo_euroc", it works well just for a while. Suddenly:
Segmentation fault (core dumped)

So I debug in gdb, and found:
Program received signal SIGSEGV, Segmentation fault.
ygz::Frame::ComputeStereoMatches (this=this@entry=0xc2403c0)
at /home/huicanlin/src/ORB-YGZ-SLAM/src/Frame.cc:671
671 const float median = vDistIdx[vDistIdx.size() / 2].first;

Any suggestions?