RGBD物体识别(2)--点云分割

对于RGBD物体识别问题，我采用的思路是先从几何特征，将物体的点云分割好，也就是将不同物体通过点云分割出来，然后通过成像关系和RGBD配准信息，在图像上找到对应的区域ROI，然后对图像进行识别，后续根据识别结果再进行抓取。当然其中还有一个很重要的地方，就是物体的姿态估计，或者说最佳抓取点的选取，这个问题暂且不管。

本文重点关注未知物体的点云分割

工具

PCL
SegmenterLight

步骤

1，安装和编译PCL1.7以上版本
我使用的时pcl1.8，到github上下载PCL压缩包，解压到pcl文件夹。

mkdir build
cd build/
sudo apt-get install cmake-curses-gui
ccmake ..

选择”BUILD_surface_on_nurbs” 为 “ON” (default “OFF”)

cmake ..
make -j4
sudo make install

2，安装依赖项

sudo apt-get install libopenni-sensor-primesense-dev 
sudo apt-get install libopenni-dev 
sudo apt-get install libopenni-sensor-primesense0 
sudo apt-get install libopenni0 

3，下载SegmenterLight
解压到pcl同级目录，如下图

cd SegmenterLight/

gedit ./v4r/CMakeLists.txt

In CMakeLists.txt , change “PCL 1.5 REQUIRED” to “PCL 1.8 REQUIRED”

mkdir build
cd build/
ccmake ..

* set “OPENNI_INCLUDE_DIRS” = “/usr/include/ni” *
* set “OPENNI_LIBRARY” = “/usr/lib/libOpenNI.so” *
* press c and then g *
注意，有可能只能找到openni2，填写对应目录和so文件即可

make -j4
sudo make install

4，测试
现在已经将其编译到系统的库目录下了，可以把这个当做一个库使用，当然，它是有demo的，但是demo的代码太过于庞杂。到SegmenterLight目录下的bin中可以找到可执行文件SegmenterLight

cd SegmenterLight/bin

执行

./SegmenterLight -h

可以看到操作说明，执行

./SegmenterLight -f learn0.pcd

learn0.pcd是放到bin下的待分割文件，出现Debug image的窗口后，点击该窗口按下F9，即可看到分割效果：

效果不错。

5，使用编译出的库
要嵌入到自己的工程中，可以直接使用编译出的动态库，
工程目录如下：

CMakeLists.txt内容

cmake_minimum_required(VERSION 2.8 FATAL_ERROR)
project(v4r_test)
find_package( OpenCV 3 REQUIRED )
find_package(PCL 1.8 REQUIRED)
include_directories(${PCL_INCLUDE_DIRS})
link_directories(${PCL_LIBRARY_DIRS})
add_definitions(${PCL_DEFINITIONS})

SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopenmp")
SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fopenmp")
SET(CMAKE_BUILD_TYPE Release)   # Mandatory for openmp!

add_executable(main main.cpp)
target_link_libraries(main ${PCL_LIBRARIES} ${OpenCV_LIBS} v4rTomGine v4rSegmenterLight)

main.cpp

#include <iostream>

#include <pcl/io/pcd_io.h>
#include <pcl/io/ply_io.h>
#include <pcl/point_cloud.h>
#include <pcl/console/parse.h>
#include <pcl/common/transforms.h>
#include <pcl/visualization/pcl_visualizer.h>
#include <pcl/visualization/cloud_viewer.h>


#include <iostream>

#include <boost/thread/thread.hpp>
#include <pcl/common/common_headers.h>
#include <pcl/features/normal_3d.h>
#include <pcl/io/pcd_io.h>
#include <pcl/visualization/pcl_visualizer.h>
#include <pcl/console/parse.h>



#include <v4r/SegmenterLight/SegmenterLight.h>

#include <opencv2/highgui/highgui.hpp>
#include <opencv2/core/core.hpp>
#include <opencv2/opencv.hpp>


#include <time.h>
using namespace std;
using namespace cv;

// This function displays the help
void showHelp(char * program_name)
{
  std::cout << std::endl;
  std::cout << "Usage: " << program_name << " cloud_filename.[pcd|ply]" << std::endl;
  std::cout << "-h:  Show this help." << std::endl;
}


void PCLCloud2Image(const pcl::PointCloud<pcl::PointXYZRGB>::Ptr & pcl_cloud, cv::Mat_<cv::Vec3b> &image){
    unsigned pcWidth = pcl_cloud->width;
    unsigned pcHeight = pcl_cloud->height;
    unsigned position = 0;

    image = cv::Mat_<cv::Vec3b>(pcHeight, pcWidth);

    for (unsigned row = 0; row < pcHeight; row++) {
      for (unsigned col = 0; col < pcWidth; col++) {
        cv::Vec3b &cvp = image.at<cv::Vec3b> (row, col);
        position = row * pcWidth + col;
        const pcl::PointXYZRGB &pt = pcl_cloud->points[position];

        cvp[2] = pt.r;
        cvp[1] = pt.g;
        cvp[0] = pt.b;
      }
    }

}

int getLabels(pcl::PointCloud<pcl::PointXYZRGBL>::Ptr cloud){
    int max_label = 0;

    set<int> result;
    for(int i=0; i<cloud->points.size(); i++){
        if(cloud->points[i].label > max_label){
            max_label = cloud->points[i].label;
        }
        result.insert(cloud->points[i].label);

    }

    cout <<"set cout begin " <<endl;
    cout <<"set.size = " << result.size()<<endl;
    for(set<int>::iterator it = result.begin(); it != result.end(); it++){
        cout << *it <<" ";
    }
    cout <<endl;

    return max_label;

}

typedef union
{
  struct
  {
    unsigned char b; // Blue channel
    unsigned char g; // Green channel
    unsigned char r; // Red channel
    unsigned char a; // Alpha channel
  };
  float float_value;
  long long_value;
} RGBValue;

inline float GetRandomColor()
{
  RGBValue x;
  x.b = std::rand()%255;
  x.g = std::rand()%255;
  x.r = std::rand()%255;
  x.a = 0.;
  return x.float_value;
}


// This is the main function
int main (int argc, char** argv)
{
    pcl::PointCloud<pcl::PointXYZRGB>::Ptr pcl_cloud(new pcl::PointCloud<pcl::PointXYZRGB>);    

    pcl::io::loadPCDFile("learn1.pcd", *pcl_cloud);
    cout << pcl_cloud->points[0].r <<endl;
    pcl::io::savePCDFileASCII("1111.txt", *pcl_cloud);
    std::string modelPath = "../model/";
    pcl::PointCloud<pcl::PointXYZRGBL>::Ptr pcl_cloud_labeled(new pcl::PointCloud<pcl::PointXYZRGBL>);
    segment::SegmenterLight seg(modelPath);
    seg.setFast(true);
    seg.setDetail(2);
    pcl_cloud_labeled = seg.processPointCloud(pcl_cloud);

    int max_label = getLabels(pcl_cloud_labeled);

    RGBValue colors[max_label+1];
    for(int i=0; i<=max_label; i++){
        colors[i].float_value = GetRandomColor();
    }

    for(int i=0; i<pcl_cloud_labeled->points.size(); i++){
        pcl::PointXYZRGB &p = pcl_cloud->points[i];
        p.rgb = colors[pcl_cloud_labeled->points[i].label].float_value;

    }

    pcl::visualization::PCLVisualizer viewer("saldfjald");

    viewer.addPointCloud(pcl_cloud);
    viewer.addCoordinateSystem(1.0, "cloud", 0);
    while(!viewer.wasStopped()){
        viewer.spinOnce();
    }
    std::cout << "okkkkk" << std::endl;
    cv::Mat_<cv::Vec3b> src;
    long int t1,t2;
    t1 = getTickCount();

    PCLCloud2Image(pcl_cloud, src);
    t2 = getTickCount();
    cout << "frequency = " << getTickFrequency() <<endl;
    cout << (t2- t1)/getTickFrequency() << " s" <<endl;

    cout << "time = " << t2 - t1 << " ms " <<endl;
    cv::imshow("saf", src);
    cv::waitKey(0);

    return 0;
}

效果：

非常不错。

好了，折腾点云分割去吧。