drew212's profile - activity

Hi, this isn't possible with RGBDSLAM at the moment (and I currently have no plans to add this feature). To implement this, in addition to the point cloud, you would also need to store the belonging feature descriptors and their 2d/3d locations. Further the graph would need to be saved (there is a function for that in hogman/g2o).

So in fact, one would need to implent saving and restoring of the Node class.

While the principle is easy, this would be a lot of programming work, and since I don't have time for it, I guess it will not be done. Anyone willing to volunteer is welcome to contact me for details.

What is AIS and where can I find further documentation on it? I'm working with RGBDSlam and it seems to use an AIS::GraphOptimizer3D* to keep track of all the clouds but I cannot find any documentation on it.

Is there a way to load the saved PCD file to rgbdslam for later use? What I am trying to do is create a map, and then later come along and locate my position/orientation within the map.

Update: I am able to load a pointcloud into a pointcloud_type for use in rgbdslam but I'm not sure how or where to add it to the graph. I have a call to setPointCloud(cloud, transform) but it isn't working correctly, as it was a shot in the dark. Below is a snippit of code, this method is set up in graph_manager.cpp, and is called from qtcv when the menu option is clicked and a file is selected. I know that it is correctly loading the file because the ROS_INFO call works and says that it is loading points.

void GraphManager::loadCloudsFromFile(QString filename){
     string * printableFileName = new string(qPrintable(filename));
     pcl::PointCloud<pcl::PointXYZRGB>::Ptr cloud(new pcl::PointCloud<pcl::PointXYZRGB>);
     if(pcl::io::loadPCDFile<pcl::PointXYZRGB>(*printableFileName, *cloud) == -1)
     {
         ROS_ERROR("Couldn't read file %s", printableFileName->c_str());
         return;
     }
     ROS_INFO("Loaded %d, data points from %s", cloud->width * cloud->height, printableFileName->c_str());
     pointcloud_type * cloud2 = new pointcloud_type(*cloud);
     //sensor_msgs::PointCloud2 cloud2;
     //pcl::toROSMsg(*cloud, cloud2);
     latest_transform_ = QMatrix4x4();
     Q_EMIT(setPointCloud(cloud2, latest_transform_));
}

In the new snapshot (see http://www.ros.org/wiki/rgbdslam/evaluation) you can store the trajectory to file. Also, if you send the model, a transformation from /map to /openni_camera is published. In the svn version, a transformation from /openni_camera to /slam_transform is published, where (somewhat confusing) /openni_camera acts as the fixed/map frame and /slam_transform is the moving camera postion.

The best way to get an accurate and complete model I have found at this time is hand correcting the errors. I use this when I need a "base mapping" that needs to be extremely accurate. Meshlab may be what you're looking for, as it allows hand modifying of point clouds in PLY and OBJ formats. It can be found here: http://meshlab.sourceforge.net/. If you're looking for an automated solution I haven't been able to get anything to work to the accuracy I need, if you find a solution let me know.

The newer openni_kinect drivers get the factory calibration settings, so calibration is no longer necessary. This is why the calibration tools are associated with the deprecated package, because it doesn't use the factory calibration settings.

I believe that the newer openni_kinect drivers get the factory calibration settings, so calibration is no longer necessary. This is why the calibration tools are associated with the deprecated package, because it doesn't use the factory calibration settings.

Is there a way to calculate or extract the camera location and pose within RGBDSlam. I've been poking at the RGBDSlam code for a while trying to figure out if data was able to be published or calculated somewhere but can't find anything. I think it is inferred somehow as the newest version prints the changed coordinate axis on each update. Thanks!

When compiling RGBDSlam I get a couple of errors. My ROS version is Electric and Ubuntu release is natty. Here are the errors:

[ 47%] Building CXX object CMakeFiles/rgbdslam.dir/src/openni_listener.o
In file included from /opt/ros/electric/stacks/rgbdslam/src/openni_listener.cpp:24:0:
/opt/ros/electric/stacks/vision_opencv/cv_bridge/include/cv_bridge/CvBridge.h: In static member function ‘static sensor_msgs::Image_<std::allocator<void> >::Ptr sensor_msgs::CvBridge::cvToImgMsg(const IplImage*, std::string)’:
/opt/ros/electric/stacks/vision_opencv/cv_bridge/include/cv_bridge/CvBridge.h:408:55: warning: ‘static bool sensor_msgs::CvBridge::fromIpltoRosImage(const IplImage*, sensor_msgs::Image&, std::string)’ is deprecated (declared at /opt/ros/electric/stacks/vision_opencv/cv_bridge/include/cv_bridge/CvBridge.h:307)
/opt/ros/electric/stacks/rgbdslam/src/openni_listener.cpp: In member function ‘void OpenNIListener::cameraCallback(const sensor_msgs::ImageConstPtr&, const sensor_msgs::ImageConstPtr&, const sensor_msgs::PointCloud2ConstPtr&)’:
/opt/ros/electric/stacks/rgbdslam/src/openni_listener.cpp:85:24: warning: ‘sensor_msgs::CvBridge::CvBridge()’ is deprecated (declared at /opt/ros/electric/stacks/vision_opencv/cv_bridge/include/cv_bridge/CvBridge.h:64)
[ 52%] Building CXX object CMakeFiles/rgbdslam.dir/src/qtcv.o
[ 57%] Building CXX object CMakeFiles/rgbdslam.dir/src/node.o
[ 63%] Building CXX object CMakeFiles/rgbdslam.dir/src/graph_manager.o
/opt/ros/electric/stacks/rgbdslam/src/graph_manager.cpp: In member function ‘void GraphManager::visualizeFeatureFlow3D(unsigned int, bool) const’:
/opt/ros/electric/stacks/rgbdslam/src/graph_manager.cpp:442:9: warning: format ‘%lu’ expects type ‘long unsigned int’, but argument 8 has type ‘std::vector<cv::DMatch, std::allocator<cv::DMatch> >::size_type’
/opt/ros/electric/stacks/rgbdslam/src/graph_manager.cpp:442:9: warning: format ‘%lu’ expects type ‘long unsigned int’, but argument 9 has type ‘std::vector<cv::DMatch, std::allocator<cv::DMatch> >::size_type’
/opt/ros/electric/stacks/rgbdslam/src/graph_manager.cpp: In function ‘void transformAndAppendPointCloud(const pointcloud_type&, pointcloud_type&, tf::Transform, float)’:
/opt/ros/electric/stacks/rgbdslam/src/graph_manager.cpp:1043:67: error: invalid conversion from ‘const float*’ to ‘float*’
/opt/ros/electric/stacks/rgbdslam/src/graph_manager.cpp:1043:67: error:   initializing argument 1 of ‘Eigen::Map<MatrixType, MapOptions, StrideType>::Map(Eigen::Map<MatrixType, MapOptions, StrideType>::PointerArgType, Eigen::Map<MatrixType, MapOptions, StrideType>::Index, Eigen::Map<MatrixType, MapOptions, StrideType>::Index, const StrideType&) [with PlainObjectType = Eigen::Matrix<float, 3, 1>, int MapOptions = 0, StrideType = Eigen::Stride<0, 0>, Eigen::Map<MatrixType, MapOptions, StrideType>::PointerArgType = float*, Eigen::Map<MatrixType, MapOptions, StrideType>::Index = int]’
make[3]: *** [CMakeFiles/rgbdslam.dir/src/graph_manager.o] Error 1