| Abstract: |  This thesis deals with 3D mapping using the Microsoft Kinect RGBD camera and a global localization in a feature based representation of the mapped environment. The properties of the Kinect and their effects on the mapping are discussed.
The Kinect is hand held at a constant height during the mapping, the only rotations are around its vertical axis.
The mapping system is inspired by the RGBD-ICP algorithm by Henry et al.[2010]. The combination of the ORB keypoints and customized BRAND is used for image feature detection and description because they present a good balance between speed and robustness. The ICP version optimized for convergence speed from Rusinkiewicz and Levoy [2001] was implemented.
The global localization system is based on the particle filter algorithm.
The features are also matched to map directly, guessing possible positions which are inserted into the particle filter set. The convergence speed is increased and the algorithm easily recovers from possible failures.
|
|---|