| Summary: | 碩士 === 國立臺灣大學 === 電機工程學研究所 === 104 === Global localization problem is one of the essential issue and is a vital part of mobile robot. Most of the service or mobile robot works in the indoor environment to accomplish household tasks. Therefore, the cognitive of environment become the necessary conditions for robot. If without understanding the environment, robots may not reach the specified position. Although GPS and Maps are great, but they only work outdoors and with clear line of sight to the sky. This issue becomes more worthy for robotic research.
This thesis describes an algorithm for localization of a robot which can efficiently estimate robot in 6 degrees-of-freedom (DoF) pose which consists of position and orientation with large scale point cloud data without giving the initial pose. We introduce the Fast Scene Recognition and Registration (FSRR) algorithm for robot 6-DoF localization in 3D point cloud map. We propose two different methods, which are FSRRv1 and FSRRv2 approach to solve the robot pose in 3D map. The FSRRv1 algorithm extract Sub-Maps descriptor by cascading several features, and learn a Distance-Metric to increase the precision of place recognition due to the environmental changes. The FSRRv2 algorithm adding image retrieval technique to improve the localization system. Both methods estimate robot pose by point set registration. Our proposed algorithms reduce the computation time needed for the point cloud registration by matching robot’s scene only with the retrieved Sub-Map in database.
Our technique has been implemented and tested extensively in different buildings. For the experimental results in FSRRv1 approach, the precision rate of scene recognition over 90% after implemented Similarity Learning. The applied Sub-Map registration was reduce the time complexity to letting robot localize itself in 3D environment more quickly. For the experimental results in FSRRv2 approach, the Coarse-Pose takes only 0.43 second per frame to estimate the possible robot pose on 3D map. The Exact-Pose implement registration refinement in order to revise the robot pose in the 3D map. The experimental results show that our Fast Scene Recognition and Registration system can localize mobile robot in a variety of large scale 3D point cloud dataset efficiently.
|
|---|
