Please use this identifier to cite or link to this item:
|Title:||Visual robot localization based on a RGB-D camera||Authors:||Gao, Jian||Keywords:||DRNTU::Engineering||Issue Date:||2014||Abstract:||RGB-D cameras present abundant data by integrating colour in formation and depth estimates. Recent years the development of low-cost sensors has led to a boom in the vision related research. Microsoft Kinect is a typical representative among them, which has been broadly exploited far beyond its initial gaming function but into many robotics laboratories. Localization is the basic and necessary step in robots navigation. Many robotic tasks request robot's location information as a must. For instance, robots work in indoor environment where GPS system could not provide data precise enough to use. Visual odometry plays a v ital role in this kind of situations. In visual odometry, images are captured by sensors and then analysed to get features. At last loads of location methods can be operated on these features to compute the actual position. Here Kinect is used as the imaging sensor to obtain visual in formation. Then the features are extracted from each image. There has been plenty of works on the 20 image feature extraction but not that much into the 30 category. In this paper, different kinds of feature descriptors are analysed and adopted including both conventional ones and those especially for RGB-D images. These features are respectively utilized in the following algorithm for the purpose of localization. Their performance could be com pared due to the accuracy of localisation.||URI:||http://hdl.handle.net/10356/64765||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||EEE Theses|
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.