On the issues of visually aided feature extraction from 3D range data

Sensing environments and data analysis are basic and important abilities that autonomous navigation systems must possess. These provide the systems capability to detect and classify objects. The advantages of working with single sensors are that the structure of the systems is less complicated, and data analysis is also less time consuming and simple. However, the sensors may not provide enough information for complicated environments, or the accuracy of the data is poor to complete certain tasks. These depend on characteristics of individual sensors and the complexity of the given tasks. This thesis provides an overview of the algorithms and software designed to extract features from 2D range data. A real time algorithm is developed to extract tree trunks, lamp posts and building pillars from the environment by using 2D laser sensors. Similarly, The Hough transform, which is applicable for range data, is developed to extract lines corresponding to building walls. These extracted landmarks are used as input in feature-based Extended Kalman Filter SLAM to improve the precision of localization of the vehicle and mapping of the environment.MASTER OF ENGINEERING (EEE