Mobile robot localization using mono vision
S V R, Krishna Vivek
Date of Issue2014
School of Electrical and Electronic Engineering
Structure from motion had been an active area of research these days. Especially vision based control and navigation is rapidly replacing traditional sensors due to low cost of cameras and their easy interfacing equipments. Localization is the key step for the robot control and navigation. In areas where GPS or other sensors could not be deployed for robot localization, visual odometry can be an alternative. Out of many visual odometry algorithms available in market, one of the most cost effective algorithm is to use a single camera to estimate the visual odometry of the robot. In this project, various algorithms for monocular visual odometry are studied. Different kinds of coordinate systems to represent motion of the robot and features in the image are identified and algorithms to implement these tasks are developed. Their performance in terms of accuracy of the visual odometry estimated from each representation is analysed. A Benchmark dataset is used to validate the algorithms. The algorithms are tested on HP webcam which is hand-held and is constraint free.