C++ 点云示例代码-Point Cloud

南宫NG·28

南宫NG·28

C++ 点云示例代码-Point Cloud

# 点云示例-PointCloud

 

功能描述:连接设备开流 ,生成深度点云或RGBD点云并保存成ply格式文件,并通过ESC_KEY键退出程序

> 本示例基于C++ High Level API进行演示

 

创建点云保存成ply格式文件函数,ply文件格式详细描述可在网络上查看<br />首先创建两个函数来保存从流里面获取到的点云数据,这是保存普通点云数据的函数

//保存点云数据到ply
void savePointsToPly(std::shared_ptr<ob::Frame> frame, std::string fileName) {
    int   pointsSize = frame->dataSize() / sizeof(OBPoint);
    FILE *fp         = fopen(fileName.c_str(), "wb+");
    fprintf(fp, "ply\n");
    fprintf(fp, "format ascii 1.0\n");
    fprintf(fp, "element vertex %d\n", pointsSize);
    fprintf(fp, "property float x\n");
    fprintf(fp, "property float y\n");
    fprintf(fp, "property float z\n");
    fprintf(fp, "end_header\n");

    OBPoint *point = (OBPoint *)frame->data();
    for(int i = 0; i < pointsSize; i++) {
        fprintf(fp, "%.3f %.3f %.3f\n", point->x, point->y, point->z);
        point++;
    }

    fflush(fp);
    fclose(fp);
}

 

再创建一个保存彩色点云数据的函数用于存储彩色点云数据

//保存彩色点云数据到ply
void saveRGBPointsToPly(std::shared_ptr<ob::Frame> frame, std::string fileName) {
    int   pointsSize = frame->dataSize() / sizeof(OBColorPoint);
    FILE *fp         = fopen(fileName.c_str(), "wb+");
    fprintf(fp, "ply\n");
    fprintf(fp, "format ascii 1.0\n");
    fprintf(fp, "element vertex %d\n", pointsSize);
    fprintf(fp, "property float x\n");
    fprintf(fp, "property float y\n");
    fprintf(fp, "property float z\n");
    fprintf(fp, "property uchar red\n");
    fprintf(fp, "property uchar green\n");
    fprintf(fp, "property uchar blue\n");
    fprintf(fp, "end_header\n");

    OBColorPoint *point = (OBColorPoint *)frame->data();
    for(int i = 0; i < pointsSize; i++) {
        fprintf(fp, "%.3f %.3f %.3f %d %d %d\n", point->x, point->y, point->z, (int)point->r, (int)point->g, (int)point->b);
        point++;
    }

    fflush(fp);
    fclose(fp);
}

 

设置Log等级,避免过多Info等级的Log影响点云输出的结果

ob::Context::setLoggerSeverity(OB_LOG_SEVERITY_ERROR);

 

创建一个Pipeline,通过Pipeline可以很容易的打开和关闭多种类型的流并获取一组帧数据

ob::Pipeline pipeline;

 

配置color流

auto colorProfiles = pipeline.getStreamProfileList(OB_SENSOR_COLOR);
if(colorProfiles) {
    auto profile = colorProfiles->getProfile(OB_PROFILE_DEFAULT);
    colorProfile = profile->as<ob::VideoStreamProfile>();
}
config->enableStream(colorProfile);

 

配置深度流

std::shared_ptr<ob::StreamProfileList> depthProfileList;
OBAlignMode                            alignMode = ALIGN_DISABLE;
if(colorProfile) {
    // Try find supported depth to color align hardware mode profile
    depthProfileList = pipeline.getD2CDepthProfileList(colorProfile, ALIGN_D2C_HW_MODE);
    if(depthProfileList->count() > 0) {
        alignMode = ALIGN_D2C_HW_MODE;
    }
    else {
        // Try find supported depth to color align software mode profile
        depthProfileList = pipeline.getD2CDepthProfileList(colorProfile, ALIGN_D2C_SW_MODE);
        if(depthProfileList->count() > 0) {
            alignMode = ALIGN_D2C_SW_MODE;
        }
    }
}
else {
    depthProfileList = pipeline.getStreamProfileList(OB_SENSOR_DEPTH);
}

if(depthProfileList->count() > 0) {
    std::shared_ptr<ob::StreamProfile> depthProfile;
    try {
        // Select the profile with the same frame rate as color.
        if(colorProfile) {
            depthProfile = depthProfileList->getVideoStreamProfile(OB_WIDTH_ANY, OB_HEIGHT_ANY, OB_FORMAT_ANY, colorProfile->fps());
        }
    }
    catch(...) {
        depthProfile = nullptr;
    }

    if(!depthProfile) {
        // If no matching profile is found, select the default profile.
        depthProfile = depthProfileList->getProfile(OB_PROFILE_DEFAULT);
    }
    config->enableStream(depthProfile);

 

开启D2C对齐, 生成RGBD点云时需要开启

// 开启D2C对齐, 生成RGBD点云时需要开启
config->setAlignMode(ALIGN_D2C_HW_MODE);

 

启动Pipeline

pipeline.start( config );

 

创建点云Filter对象,并且设置相机内参

// 创建点云Filter对象(点云Filter创建时会在Pipeline内部获取设备参数, 所以尽量在Filter创建前配置好设备)
ob::PointCloudFilter pointCloud;

//获取相机内参传入点云Filter中
auto cameraParam = pipeline.getCameraParam();
pointCloud.setCameraParam(cameraParam)

 

设置些操作提示

 std::cout << "Press R to create rgbd pointCloud and save to ply file! " << std::endl;
 std::cout << "Press d to create depth pointCloud and save to ply file! " << std::endl;
 std::cout << "Press ESC to exit! " << std::endl;

设置主流程通过上面创建的点云Filter对象获取并保存点云数据

if(key == 'R' || key == 'r') {
  count = 0;
  //限制最多重复10次
  while(count++ < 10) {
    //等待一帧数据,超时时间为100ms
    auto frameset = pipeline.waitForFrames(100);
    if(frameset != nullptr && frameset->depthFrame() != nullptr && frameset->colorFrame() != nullptr) {
      try {
        //生成彩色点云并保存
        std::cout << "Save RGBD PointCloud ply file..." << std::endl;
        pointCloud.setCreatePointFormat(OB_FORMAT_RGB_POINT);
        std::shared_ptr<ob::Frame> frame = pointCloud.process(frameset);
        saveRGBPointsToPly(frame, "RGBPoints.ply");
        std::cout << "RGBPoints.ply Saved" << std::endl;
      }
      catch(std::exception &e) {
        std::cout << "Get point cloud failed" << std::endl;
      }
      break;
    }
  }
}
else if(key == 'D' || key == 'd') {
  count = 0;
  //限制最多重复10次
  while(count++ < 10) {
    //等待一帧数据,超时时间为100ms
    auto frameset = pipeline.waitForFrames(100);
    if(frameset != nullptr && frameset->depthFrame() != nullptr) {
      try {
        //生成点云并保存
        std::cout << "Save Depth PointCloud to ply file..." << std::endl;
        pointCloud.setCreatePointFormat(OB_FORMAT_POINT);
        std::shared_ptr<ob::Frame> frame = pointCloud.process(frameset);
        savePointsToPly(frame, "DepthPoints.ply");
        std::cout << "DepthPoints.ply Saved" << std::endl;
      }
      catch(std::exception &e) {
        std::cout << "Get point cloud failed" << std::endl;
      }
      break;
    }
  }
}

 

最后通过Pipeline来停止流

 pipeline.stop();

 

程序正常退出后会释放资源

预期输出:

image.png 


友情链接: