基于ORB-SLAM2实时构建稠密点云

本篇博客最早发布于实验室公共博客，但已无人维护，现迁移至个人博客

---

ORB-SLAM2是特征点法的经典之作，但是只能构建稀疏二维点云，限制了其使用范围

因期望可以实现移动机器人的导航功能，需要构建三维点云，再通过octomap_server转化为八叉树地图或者栅格地图，运用到实际中去，因此需要在原代码上做些改动

稠密点云的重建工作高翔博士已经完成了一些，感谢高博！我的改进是基于它做的

ORBSLAM2_with_pointcloud_map安装

首先Git下

https://github.com/gaoxiang12/ORBSLAM2_with_pointcloud_map

安装的时候大家可能会遇到一些问题，幸好已经有大佬帮大家做了整理

安装指导

https://blog.csdn.net/Xuesengxinyi/article/details/93469079?ops_request_misc=%257B%2522request%255Fid%2522%253A%2522164968342916782094899573%2522%252C%2522scm%2522%253A%252220140713.130102334.pc%255Fall.%2522%257D&request_id=164968342916782094899573&biz_id=0&utm_medium=distribute.pc_search_result.none-task-blog-2_allfirst_rank_ecpm_v1~rank_v31_ecpm-4-93469079.142^v7pc_search_result_control_group,157^v4control&utm_term=orb+slam保存为pcd&spm=1018.2226.3001.4187

安装问题总结

https://blog.csdn.net/c417469898/article/details/104631814

建议大家先按着安装指导走一遍，有问题再去看安装问题总结

大家可能需要卸载高版本的Eigen，Eigen的卸载和安装可以看下面的博客

https://blog.csdn.net/qq_45401419/article/details/118358687

ORB_SLAM2_with_pointcloud编译报错

../lib/libORB_SLAM2.so：对‘DBoW2::FORB::fromString(cv::Mat&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)’未定义的引用
../lib/libORB_SLAM2.so：对‘DBoW2::FORB::toString[abi:cxx11](cv::Mat const&)’未定义的引用

参考解决：

https://blog.csdn.net/chishuideyu/article/details/77189561

Kinect V1标定与yaml文件编写

实验室暂时只有Kinect V1的RGBD相机，虽然不是最新的产品，但是有的玩就行

ubuntu16.04安装微软kinect V1驱动

主要参考：

https://blog.csdn.net/hongge_smile/article/details/107856087

./RedistMaker的时候报错：

/usr/include/ni/XnOS.h:327:49: error: cannot allocate an object of abstract type 'XnSensorDepthGenerator'
  #define XN_NEW(type, ...)  new type(__VA_ARGS__)

参考网上的一些资料：

https://stackoverflow.com/questions/36601607/c-abstract-error

https://github.com/avin2/SensorKinect/pull/5

在XnSensorGenerator.h和XnSensorDepthGenerator.h文件中添加：

XnStatus GetPixelCoordinatesInViewPoint(xn::ProductionNode& other, XnUInt32 x, XnUInt32 y, XnUInt32& altX, XnUInt32& altY)
{
    return 0;
}

再次./RedistMaker，成功运行，打开终端，输入：

roslaunch freenect_launch freenect.launch

打开rviz，Add→by topic，添加后可查看图像，说明RGBD相机启动成功

Kinect V1相机内参标定

http://wiki.ros.org/openni_launch/Tutorials/IntrinsicCalibration

https://blog.csdn.net/puqian13/article/details/103163585

标定完成后再~/.ros目录下会有camera_info子目录，里面存放着标定完成后的yaml文件

但是该yaml文件和ORB-SLAM2需要的参数设置文件还是有出入的

ORB-SLAM源码中的相机参数设置

https://blog.csdn.net/catpico/article/details/120688795

注意相机内参和畸变系数的对应值

但是ORB-SLAM2设置文件中有几个参数还是有待商榷的，主要是以下两个

• 尺度因子(DepthScaleFactor)
• Camera.bf

有以下讨论：

http://www.luohanjie.com/2017-04-05/the-problem-of-calibration-data-in-orb-slam2.html

https://github.com/raulmur/ORB_SLAM2/issues/737

Camera.bf是fx乘baseline，Kinect的baseline参照上面应该是7.5cm

Kinect V1构造

http://wiki.ros.org/kinect_calibration/technical

点云地图与八叉树地图保存

高翔博士实现了点云地图的显示功能，我在其基础上增加了3个功能：

• 点云地图的保存
• 八叉树地图的绘制与保存
• rviz实时增量式显示稠密点云

这一节先讲前两个

在ponitcloudmapping.h中添加头文件：

#include <pcl-1.7/pcl/io/pcd_io.h>
#include <octomap/octomap.h>
#include <octovis/OcTreeDrawer.h>

添加八叉树点云生成函数

octomap::Pointcloud generateOctoCloud(KeyFrame* kf, cv::Mat& color, cv::Mat& depth);

在viewer函数中完成具体的功能

void PointCloudMapping::viewer()
{
    pcl::visualization::CloudViewer viewer("viewer");
    octomap::OcTree tree(0.05);

    while(1)
    {
        {
            unique_lock<mutex> lck_shutdown( shutDownMutex );
            if (shutDownFlag)
            {
                break;
            }
        }
        {
            unique_lock<mutex> lck_keyframeUpdated( keyFrameUpdateMutex );
            keyFrameUpdated.wait( lck_keyframeUpdated );
        }
        
        // keyframe is updated 
        size_t N=0;
        {
            unique_lock<mutex> lck( keyframeMutex );
            N = keyframes.size();
        }
        
        for ( size_t i=lastKeyframeSize; i<N ; i++ )
        {
            PointCloud p = generatePointCloud( keyframes[i], colorImgs[i], depthImgs[i] );
            octomap::Pointcloud octo_p = generateOctoCloud( keyframes[i], colorImgs[i], depthImgs[i] );

            globalMap += p;
            Isometry3d Tcw = ORB_SLAM2::Converter::toSE3Quat( keyframes[i]->GetPose());
            Isometry3d Twc = Tcw.inverse();
            tree.insertPointCloud(octo_p,octomap::point3d( Twc(0,3), Twc(1,3), Twc(2,3) ));        
        }
        
        string map_path="/home/redwall/ORB_SLAM2_modified/map";

        // 更新中间节点的占据信息并写入磁盘
        tree.updateInnerOccupancy();
        cout<<"saving octomap ... "<<endl;
        tree.writeBinary( map_path+"/octomap/octomap.bt" );

        PointCloud tmp;
        PointCloud::Ptr cloudPointer(new PointCloud);
        cloudPointer = globalMap.makeShared();
        cout<<"cloudPointer size:"<<cloudPointer->size()<<endl;
        voxel.setInputCloud( cloudPointer );        
        voxel.filter( tmp );
        globalMap.swap( tmp );

        cout<<"saving global map ..."<<endl;
        pcl::io::savePCDFileBinary(map_path+"/pointcloud/map.pcd",globalMap);
        viewer.showCloud( cloudPointer );
        cout<<"show global map, size="<<globalMap.points.size()<<endl;        

        lastKeyframeSize = N;
    }
}

实际运行，地图保存至目标位置，我们分别用pcl_viewer和octovis查看，效果如下：

其中运行octovis的时候报错

octovis: error while loading shared libraries: libQGLViewer.so.2: cannot open shared object file: No such file or directory

1、删除octomap和octovis目录下的build文件夹

2、在octomap_devel目录下新建build文件夹，并编译安装

3、locate libQGLViewer.so.2

4、将路径下的libQGLViewer.so*复制到/usr/local/lib目录下

问题解决，顺利查看octomap地图

rviz实时增量式显示稠密点云

这部分折磨了我好几天，索性最后解决了，其实博客上已经有许多大佬给出了实现，比如博主熊猫飞天，我也阅读了其博客，写的还是很好的，值得学习

我的实现相较于熊猫飞天稍微简单些，也较易于理解

1版

1版是个失败的版本，大家时间赶的话可以直接跳过

编写测试程序如下：

int main(int argc, char **argv)
{
    ros::init(argc, argv, "RGBD");
    ros::start();

    if(argc != 3)
    {
        cerr << endl << "Usage: rosrun ORB_SLAM2 RGBD path_to_vocabulary path_to_settings" << endl;        
        ros::shutdown();
        return 1;
    }    

    // Create SLAM system. It initializes all system threads and gets ready to process frames.
    ORB_SLAM2::System SLAM(argv[1],argv[2],ORB_SLAM2::System::RGBD,true);

    ImageGrabber igb(&SLAM);

    ros::NodeHandle nh;

    message_filters::Subscriber<sensor_msgs::Image> rgb_sub(nh, "/camera/rgb/image_color", 1);
    message_filters::Subscriber<sensor_msgs::Image> depth_sub(nh, "/camera/depth_registered/image_raw", 1);
    typedef message_filters::sync_policies::ApproximateTime<sensor_msgs::Image, sensor_msgs::Image> sync_pol;
    message_filters::Synchronizer<sync_pol> sync(sync_pol(10), rgb_sub,depth_sub);
    sync.registerCallback(boost::bind(&ImageGrabber::GrabRGBD,&igb,_1,_2));

    ros::Publisher pcl_pub = nh.advertise<sensor_msgs::PointCloud2>("/pointcloud_output",10);

    sensor_msgs::PointCloud2 output;
    pcl::PointCloud<pcl::PointXYZRGB> cloud(*SLAM.mpPointCloudMapping->globalMap);
    // cout<<"mpTracker globalMap size:"<<SLAM.mpTracker->mpPointCloudMapping->globalMap->size()<<endl;
    cout<<"system globalMap size:"<<SLAM.mpPointCloudMapping->globalMap->size()<<endl;
    // cloud=*SLAM.mpPointCloudMapping->globalMap;
    cout<<"cloud size:"<<cloud.size()<<endl;

    cout<<"Transform globalMap...."<<endl;
    pcl::toROSMsg(*SLAM.mpPointCloudMapping->globalMap,output);
    cout<<"output pointclud height:"<<output.height<<endl;
    cout<<"output pointcloud width:"<<output.width<<endl;
    cout<<"----------------------"<<endl;
    
    output.header.stamp = ros::Time::now();
    output.header.frame_id = "world";

    double dt=0.1;
    ros::Rate loop_rate(1/dt);
    while(ros::ok())
    {
        // cout<<"mpTracker globalMap size:"<<SLAM.mpTracker->mpPointCloudMapping->globalMap->size()<<endl;
        cout<<"system globalMap size:"<<SLAM.mpPointCloudMapping->globalMap->size()<<endl;
        cout<<"cloud size:"<<cloud.size()<<endl;
        cout<<"output pointclud height:"<<output.height<<endl;
        cout<<"output pointcloud width:"<<output.width<<endl<<endl;

        pcl_pub.publish(output);
        ros::spinOnce();
        loop_rate.sleep();
    }
    
    // ros::spin();

    // Stop all threads
    SLAM.Shutdown();

    // Save camera trajectory
    SLAM.SaveKeyFrameTrajectoryTUM("KeyFrameTrajectory.txt");

    ros::shutdown();

    return 0;
}

pcl::PointCloud转化为sensor_msgs::PointCloud2后，实际sensor_msgs::PointCloud2中并没有数据

同时用*SLAM.mpPointCloudMapping->globalMap初始化同类型的cloud后，cloud的size也是0

运行截图如下：

但是地图是可以正常保存的(在pointcloudmapping.cc中更改了地图保存的位置)

编写地图加载与发布rviz话题的测试程序，如下：

#include <iostream>
#include <sensor_msgs/PointCloud2.h>
#include <pcl_conversions/pcl_conversions.h>
#include <pcl-1.7/pcl/io/pcd_io.h>

#include <ros/ros.h>

using namespace std;

int main(int argc,char** argv)
{
    ros::init(argc,argv,"PointCloud");
    ros::NodeHandle nh;

    ros::Publisher pcl_pub = nh.advertise<sensor_msgs::PointCloud2>("/pointcloud_output",10);

    pcl::PointCloud<pcl::PointXYZRGB> cloud;
    sensor_msgs::PointCloud2 output;
    pcl::io::loadPCDFile("/home/redwall/ORB_SLAM2_modified/map/pointcloud/map.pcd",cloud);
    pcl::toROSMsg(cloud,output);
    
    output.header.stamp = ros::Time::now();
    output.header.frame_id = "world";

    double dt=0.1;
    ros::Rate loop_rate(1/dt);
    while(ros::ok())
    {
        cout<<"output pointclud height:"<<output.height<<endl;
        cout<<"output pointcloud width:"<<output.width<<endl<<endl;
        pcl_pub.publish(output);
        ros::spinOnce();
        loop_rate.sleep();
    }

    return 0;
}

运行正常且可在Rviz中显示

并且global map的size应等于output pointcloud width

💡 width是点云的长度

令人头秃

2版

好吧，我实在是有些愚蠢，也实在粗心

我一开始以为是shared_ptr指针的问题，所以把代码里的所有智能指针都替换成了普通指针，结果自然是白费功夫，当然唯一的好处是代码的整洁性有所提高(手动狗头)

//Initialize the Tracking thread
//(it will live in the main thread of execution, the one that called this constructor)
mpPointCloudMapping = new PointCloudMapping( resolution );
mpTracker = new Tracking(this, mpVocabulary, mpFrameDrawer, mpMapDrawer,
                          mpMap, mpPointCloudMapping, mpKeyFrameDatabase, strSettingsFile, mSensor);

Tracking::Tracking(System *pSys, ORBVocabulary* pVoc, FrameDrawer *pFrameDrawer, MapDrawer *pMapDrawer, Map *pMap, PointCloudMapping* pPointCloud, KeyFrameDatabase* pKFDB, const string &strSettingPath, const int sensor):
    mState(NO_IMAGES_YET), mSensor(sensor), mbOnlyTracking(false), mbVO(false), mpORBVocabulary(pVoc),
    mpPointCloudMapping( pPointCloud ),
    mpKeyFrameDB(pKFDB), mpInitializer(static_cast<Initializer*>(NULL)), mpSystem(pSys),
    mpFrameDrawer(pFrameDrawer), mpMapDrawer(pMapDrawer), mpMap(pMap), mnLastRelocFrameId(0)
{
    // Load camera parameters from settings file

言归正传，我仔细研究了一下终端的输出，发现下面这段代码实际上只在SLAM系统初始化的时候执行了一次

		cout<<"Transform globalMap...."<<endl;
    pcl::toROSMsg(*SLAM.mpPointCloudMapping->globalMap,output);
    cout<<"output pointclud height:"<<output.height<<endl;
    cout<<"output pointcloud width:"<<output.width<<endl;
    cout<<"----------------------"<<endl;

所以每次发布点云的width都是0，难怪rviz中啥都没有

在研究一下发现图像的处理都是在回调函数void ImageGrabber::GrabRGBD( )中完成

那么在回调函数中转化点云数据格式是否可行呢？

试了一下可以正常运行，实际运行中会出现fixed frame不存在的问题，运行tf的静态变换可以解决问题，参考如下：

https://blog.csdn.net/xu_fengyu/article/details/86562827

https://blog.csdn.net/weixin_44827364/article/details/104190513

https://stackoverflow.com/questions/52420672/ros-rviz-how-to-visualize-a-point-cloud-that-doesnt-have-a-fixed-frame-transfo

实际效果如下：

（文件太大embed不了…）

roslaunch问题

为了避免每次运行程序都要在终端输上很长一串，费时还容易出错，所以我写了个launch文件，如下：

<launch>

    <!-- <include file="$(find freenect_launch)/launch/examples/freenect-registered-xyzrgb.launch" /> -->

    <arg name="vocabulary_path" default="/home/redwall/ORB_SLAM2/Vocabulary/ORBvoc.txt"/>

    <arg name="settings_path" default="/home/redwall/ORB_SLAM2_modified/Examples/ROS/ORB_SLAM2/my_yaml.yaml"/>

    <node pkg="ORB_SLAM2" type="RGBD" name="ORB_ROS_RGBD" output="screen" args="$(arg vocabulary_path) $(arg settings_path)" />
        
    <!-- <node pkg="rviz" type="rviz" name="rviz"/>             -->

</launch>

在终端roslaunch上面编写的launch文件，Ctrl+C之后，目录下没有保存点云地图和八叉树地图

而用rosrun命令，则可以保存两种地图

同时修改之后，没有保存“KeyFrameTrajectory.txt”文件

这说明while循环后的语句都没有执行，关键的SLAM.Shutdown( )也没有执行

这说明没有等待各个线程完成，SLAM就终止了，当然也不会去进行全局BA，这可能会影响建图的精度，这是后面需要继续研究和改进的