rgbdslam: SLAM on RGBD Data
This package can be used to register the point clouds from RGBD sensors such as the kinect or stereo cameras.
The rgbdslam node can be connected easily to an octomap_server node to create a memory-efficient 3D map.
RGBD-SLAM Implements a SLAM-Frontend based on structure from motion using visual features to identify keypoints in the RGBD image of a kinect. It makes use of the g2o SLAM backend to compute a globally consistent trajectory from the individual transformations provided by the frontend.
The main classes are the following:
Node - Holds the data for a graph node, e.g., the extracted features and their position in the image plane and the local 3D coordinate systems. It provides functionality to find the transformation to another node and for publishing the associated pointcloud.
GraphManager - Decides for each Node, which edges are included in the graph. Stores all Nodes with valid edges and optimizes the camera pose graph globally with HOG-MAN as SLAM backend. Provides functionality to send (ROS-Msgs) or save (to PCD File) the globally consistent whole model.
The following classes are mainly for communication with ROS and the user:
OpenNIListener - Subscribes to the openni topics, constructs a node for each image-pointcloud pair and hands it to the graph manager. Online visualization results are sent out.
Graphical_UI - Constructs a QT GUI for easy control of the program
RosUi - a UI for controlling the program via ROS service calls
QtROS - Sets up a thread for ROS event processing, to seperate SLAM-computations from the GUI
GLViewer - OpenGL based display of the 3d model