原文地址http://blog.csdn.net/aptx704610875/article/details/51490201 支持原创,感谢半闲居士和易科实验室大牛
最近写完了windows上的实时rgbd_slam后,读了些论文,想着怎么改进程序,想在闭环检测的方面尝试一下。最近很火的ORB_SLAM2使用了DBoW2(ORB词袋)的方法,极大的提高了速度和匹配准确度,windows版的orb_slam2还没跑成功(一部分库的编译出现了问题,不过等研究做完了,会继续跑windows版本的),这几天一直在尝试ubuntu版的orb_slam的实时重建,今天终于成功了!~(感谢高博士为我们提供了加了3D建图模块的libORB_SLAM2.so(高博的博客:半闲居士))
首先orb_slam2的话,github下载源码编译很容易,按照官方github下面的教程走就行。晒几张TUM数据集的结果:
desk:
room:
效果很棒,模拟轨迹和groundtruth的绝对误差真的和论文上说的一样小。我觉得ORB_SLAM2真的是现在视觉SLAM里最优秀的一版,考虑的非常全面。
那么如果我们要用到自己的项目中,该怎么调用呢?特别棒的一点是,原作者提供了libORB_SLAM2.so给我们,加上头文件System.h,我们就可以把ORB_SLAM作为一个整体加到我们的项目中。但是源码中并没有3D建图的模块,需要做相应改变,高博士为我们提供了加了3D建图模块的libORB_SLAM2.so,这时我们就可以根据自己的需求(kinect,xtion或其他可以获得点云的sensors)。高博的博客中有一篇是用的kinect2,在ROS运行的orb_slam2,今天我们来试一试不用ROS,通过OpenNI2直接调用xtion获取rgb数据和depth数据来重建环境(之前有windows上运行openni2_xtion的经验)。我的建议,要么xtion要么kinect2, 因为kinect很鸡肋,xtion比它轻巧,kinect2比它分辨率高。(源码下载csdn 源码下载github)(词袋文件太大,各位可以从官方github下载)
扯了这么多,现在拉回主线。在ORB_SLAM2/Examples/RGB-D/中,创建rgbd_xtion_cc.cpp:
- #include <iostream>
- #include <algorithm>
- #include <fstream>
- #include <chrono>
- #include <OpenNI.h>
- #include <opencv2/core/core.hpp>
- #include <opencv2/highgui/highgui.hpp>
- #include <opencv2/imgproc/imgproc.hpp>
-
- #include <System.h> // orb_slam2
-
- using namespace std;
- using namespace openni;
- using namespace cv;
-
- void showdevice(){
-
- Array<DeviceInfo> aDeviceList;
- OpenNI::enumerateDevices(&aDeviceList);
-
- cout << "电脑上连接着 " << aDeviceList.getSize() << " 个体感设备." << endl;
-
- for (int i = 0; i < aDeviceList.getSize(); ++i)
- {
- cout << "设备 " << i << endl;
- const DeviceInfo& rDevInfo = aDeviceList[i];
- cout << "设备名: " << rDevInfo.getName() << endl;
- cout << "设备Id: " << rDevInfo.getUsbProductId() << endl;
- cout << "供应商名: " << rDevInfo.getVendor() << endl;
- cout << "供应商Id: " << rDevInfo.getUsbVendorId() << endl;
- cout << "设备URI: " << rDevInfo.getUri() << endl;
-
- }
- }
-
- Status initstream(Status& rc, Device& xtion, VideoStream& streamDepth, VideoStream& streamColor)
- {
- rc = STATUS_OK;
-
-
- rc = streamDepth.create(xtion, SENSOR_DEPTH);
- if (rc == STATUS_OK)
- {
-
- VideoMode mModeDepth;
-
- mModeDepth.setResolution(640, 480);
-
- mModeDepth.setFps(30);
-
- mModeDepth.setPixelFormat(PIXEL_FORMAT_DEPTH_1_MM);
-
- streamDepth.setVideoMode(mModeDepth);
- streamDepth.setMirroringEnabled(false);
-
-
- rc = streamDepth.start();
- if (rc != STATUS_OK)
- {
- cerr << "无法打开深度数据流:" << OpenNI::getExtendedError() << endl;
- streamDepth.destroy();
- }
- }
- else
- {
- cerr << "无法创建深度数据流:" << OpenNI::getExtendedError() << endl;
- }
-
-
- rc = streamColor.create(xtion, SENSOR_COLOR);
- if (rc == STATUS_OK)
- {
-
- VideoMode mModeColor;
- mModeColor.setResolution(640, 480);
- mModeColor.setFps(30);
- mModeColor.setPixelFormat(PIXEL_FORMAT_RGB888);
-
- streamColor.setVideoMode(mModeColor);
- streamColor.setMirroringEnabled(false);
-
- rc = streamColor.start();
- if (rc != STATUS_OK)
- {
- cerr << "无法打开彩色图像数据流:" << OpenNI::getExtendedError() << endl;
- streamColor.destroy();
- }
- }
- else
- {
- cerr << "无法创建彩色图像数据流:" << OpenNI::getExtendedError() << endl;
- }
-
- if (!streamColor.isValid() || !streamDepth.isValid())
- {
- cerr << "彩色或深度数据流不合法" << endl;
- OpenNI::shutdown();
- rc = STATUS_ERROR;
- return rc;
- }
-
-
- if (xtion.isImageRegistrationModeSupported(
- IMAGE_REGISTRATION_DEPTH_TO_COLOR))
- {
- xtion.setImageRegistrationMode(IMAGE_REGISTRATION_DEPTH_TO_COLOR);
- }
-
- return rc;
- }
-
- int main(int argc, char **argv)
- {
- if(argc != 3)
- {
- cerr << endl << "Usage: ./rgbd_cc path_to_vocabulary path_to_settings" << endl;
- return 1;
- }
-
-
- ORB_SLAM2::System SLAM(argv[1],argv[2],ORB_SLAM2::System::RGBD,true);
-
- cout << endl << "-------" << endl;
- cout << "Openning Xtion ..." << endl;
-
- Status rc = STATUS_OK;
-
- OpenNI::initialize();
- showdevice();
-
- Device xtion;
- const char * deviceURL = openni::ANY_DEVICE;
- rc = xtion.open(deviceURL);
-
- VideoStream streamDepth;
- VideoStream streamColor;
- if(initstream(rc, xtion, streamDepth, streamColor) == STATUS_OK)
- cout << "Open Xtion Successfully!"<<endl;
- else
- {
- cout << "Open Xtion Failed!"<<endl;
- return 0;
- }
-
-
- cv::Mat imRGB, imD;
- bool continueornot = true;
-
- VideoFrameRef frameDepth;
- VideoFrameRef frameColor;
- namedWindow("RGB Image", CV_WINDOW_AUTOSIZE);
- for (double index = 1.0; continueornot; index+=1.0)
- {
- rc = streamDepth.readFrame(&frameDepth);
- if (rc == STATUS_OK)
- {
- imD = cv::Mat(frameDepth.getHeight(), frameDepth.getWidth(), CV_16UC1, (void*)frameDepth.getData());
- }
- rc = streamColor.readFrame(&frameColor);
- if (rc == STATUS_OK)
- {
- const Mat tImageRGB(frameColor.getHeight(), frameColor.getWidth(), CV_8UC3, (void*)frameColor.getData());
- cvtColor(tImageRGB, imRGB, CV_RGB2BGR);
- imshow("RGB Image",imRGB);
- }
- SLAM.TrackRGBD( imRGB, imD, index);
- char c = cv::waitKey(5);
- switch(c)
- {
- case 'q':
- case 27:
- continueornot = false;
- break;
- case 'p':
- cv::waitKey(0);
- break;
- default:
- break;
- }
- }
-
- SLAM.Shutdown();
- SLAM.SaveTrajectoryTUM("trajectory.txt");
- cv::destroyAllWindows();
- return 0;
- }
思路很简单,首先创建orb_slam系统,传入词袋和xtion/orb参数; 然后从xtion得到彩色图和深度图,调用slam的tracking线程处理得到位姿(当然也有loop线程的闭环检测和g2o下线程的优化),融合点云到同一个坐标下并显示(pointcloudmapping.h / cc里有声明和定义)。
然后在ORB_SLAM2/CMakeLists中添加:
- find_package(OpenNI2 REQUIRED)
- include_directories("/usr/include/openni2/")
- LINK_LIBRARIES( ${OpenNI2_LIBRARY} )
- target_link_libraries(${PROJECT_NAME}${OpenNI2_LIBRARY})
- add_executable(rgbd_xtion_cc
- Examples/RGB-D/rgbd_xtion_cc.cpp)
- target_link_libraries(rgbd_xtion_cc ${PROJECT_NAME})
然后就可以在ORB_SLAM2/build/里cmake .. 和 make了。完成后可以看到ORB_SLAM2/Examples/RGB-D/里有可执行文件rgbd_xtion_cc。(rgbd_tum是跑TUM数据集的, rgbd_cc是跑自己的数据集的,这两个都是预先采集好彩色图和深度图)
最后在ORB_SLAM2/Examples/RGB-D/里创建xtion的参数文件xtion.yaml (包含了ORB参数信息),大家根据标定(OpenCV,ROS,MATLAB等)结果自行修改内参(rgb内参和畸变):
- %YAML:1.0
-
- #--------------------------------------------------------------------------------------------
- # Camera Parameters. xtion 640*480
- #--------------------------------------------------------------------------------------------
-
- # Camera calibration and distortion parameters (OpenCV)
- Camera.fx: 558.341390
- Camera.fy: 558.387543
- Camera.cx: 314.763671
- Camera.cy: 240.992295
-
- Camera.k1: 0.062568
- Camera.k2: -0.096148
- Camera.p1: 0.000140
- Camera.p2: -0.006248
- Camera.k3: 0.000000
-
- Camera.width: 640
- Camera.height: 480
-
- # Camera frames per second
- Camera.fps: 30.0
-
- # IR projector baseline times fx (aprox.)
- Camera.bf: 40.0
-
- # Color order of the images (0: BGR, 1: RGB. It is ignored if images are grayscale)
- Camera.RGB: 0
-
- # Close/Far threshold. Baseline times.
- ThDepth: 50.0
-
- # Deptmap values factor
- DepthMapFactor: 1000.0
-
- #--------------------------------------------------------------------------------------------
- # ORB Parameters
- #--------------------------------------------------------------------------------------------
-
- # ORB Extractor: Number of features per image
- ORBextractor.nFeatures: 1000
-
- # ORB Extractor: Scale factor between levels in the scale pyramid
- ORBextractor.scaleFactor: 1.2
-
- # ORB Extractor: Number of levels in the scale pyramid
- ORBextractor.nLevels: 8
-
- # ORB Extractor: Fast threshold
- # Image is divided in a grid. At each cell FAST are extracted imposing a minimum response.
- # Firstly we impose iniThFAST. If no corners are detected we impose a lower value minThFAST
- # You can lower these values if your images have low contrast
- ORBextractor.iniThFAST: 20
- ORBextractor.minThFAST: 7
-
- #--------------------------------------------------------------------------------------------
- # Viewer Parameters
- #--------------------------------------------------------------------------------------------
- Viewer.KeyFrameSize: 0.05
- Viewer.KeyFrameLineWidth: 1
- Viewer.GraphLineWidth: 0.9
- Viewer.PointSize:2
- Viewer.CameraSize: 0.08
- Viewer.CameraLineWidth: 3
- Viewer.ViewpointX: 0
- Viewer.ViewpointY: -0.7
- Viewer.ViewpointZ: -1.8
- Viewer.ViewpointF: 500
-
- #--------------------------------------------------------------------------------------------
- # PointCloud Mapping
- #--------------------------------------------------------------------------------------------
- PointCloudMapping.Resolution: 0.01
现在来运行吧~ 在ORB_SLAM2/下打开终端,输入 ./Examples/RGB-D/rgbd_xtion_cc Vocabulary/ORBvoc.txt Examples/RGB-D/xtion.yaml 系统加载ORB词袋,然后打开xtion设备,采集图像处理,显示角点,轨迹和点云:
按‘q’或esc程序退出,自动保存估计的轨迹和点云pcd文件到ORB_SLAM2/下(帧数较多时如3000帧,保存时间较长20s左右)。运行pcl_viewer xx.pcd 即可查看。保存优化后的点云的代码在pointcloudmapping.cc里:
- globalMap->clear();
- for(size_t i=0;i<keyframes.size();i++)
- {
- cout<<"keyframe "<<i<<" ..."<<endl;
- PointCloud::Ptr tp = generatePointCloud( keyframes[i], colorImgs[i], depthImgs[i] );
- PointCloud::Ptr tmp(new PointCloud());
- voxel.setInputCloud( tp );
- voxel.filter( *tmp );
- *globalMap += *tmp;
- viewer.showCloud( globalMap );
- }
- PointCloud::Ptr tmp(new PointCloud());
- sor.setInputCloud(globalMap);
- sor.filter(*tmp);
- globalMap->swap( *tmp );
- pcl::io::savePCDFileBinary ( "optimized_pointcloud.pcd", *globalMap );
- cout<<"Save point cloud file successfully!"<<endl;
局部1:
局部2:
局部3:
接下来准备改进词袋,尝试加入3d特征描述words,训练然后提高匹配精准度,拭目以待。
///
2016/6/23补充:
很多同学没有FindOpenNI2.cmake, 导致了系统找不到openni2的库
这里给大家提供了一个FindOpenNI2.cmake文件,复制内容到新的cmake文件,
保存后存到 /usr/share/cmake-2.8/Modules/中去就好了。
- #
- # Try to find OPenNI2 library and include path.
- # Once done this will define
- #
- #
-
- FIND_PATH( OpenNI2_INCLUDE_PATH OpenNI.h
- /usr/include
- /usr/local/include
- /sw/include
- /opt/local/include
- DOC "The directory where OpenNI.h resides")
- FIND_LIBRARY( OpenNI2_LIBRARY
- NAMES OpenNI2 openni2
- PATHS
- /usr/lib64
- /usr/lib
- /usr/local/lib64
- /usr/local/lib
- /sw/lib
- /opt/local/lib
- DOC "The OpenNI2 library")
-
- IF (OpenNI2_INCLUDE_PATH)
- SET( OpenNI2_FOUND 1 CACHE STRING "Set to 1 if OpenNI2 is found, 0 otherwise")
- ELSE (OpenNI2_INCLUDE_PATH)
- SET( OpenNI2_FOUND 0 CACHE STRING "Set to 1 if OpenNI2 is found, 0 otherwise")
- ENDIF (OpenNI2_INCLUDE_PATH)
-
- MARK_AS_ADVANCED( OpenNI2_FOUND )
///
2016/8/12补充:
(回答4楼的问题,由于CSDN回复功能的渣过滤技术,只能写在这里了)
1。Camera.bf中的b指基线baseline(单位:米),f是焦距fx(x轴和y轴差距不大),bf=b*f,和ThDepth一起决定了深度点的范围:bf * ThDepth / fx即大致为b * ThDepth。 基线在双目视觉中出现的比较多,如ORB-SLAM中的双目示例中的EuRoC.yaml中的bf为47.9,ThDepth为35,fx为435.2,则有效深度为47.9*35/435.3=3.85米;KITTI.yaml中的bf为387.57,ThDepth为40,fx为721.54,则有效深度为387.57*40/721.54=21.5米。这里的xtion的IR基线(其实也可以不这么叫)bf为40,ThDepth为50,fx为558.34,则有效深度为3.58米(官方为3.5米)。
2。DepthMapFactor: 1000.0这个很好理解,depth深度图的值为真实3d点深度*1000. 例如depth值为2683,则真是世界尺度的这点的深度为2.683米。 这个值是可以人为转换的(如opencv中的convert函数,可以设置缩放因子),如TUM中的深度图的DepthMapFactor为5000,就代表深度图中的5000个单位为1米。
未来,属于一心想要改变世界的人。
-cc