Texture and Distinctness Analysis for Natural Feature Extraction

摘要

One of the basic requirements for autonomous navigation in an unexplored and often complex environment is to be able to lock on to natural features. This paper presents a method for extracting features distinctive enough to navigate with. The method consists of three parts. Firstly, it selects a set of interest points from the images which are invariant to most changes in conditions; secondly, it analyses the texture distribution of the local interest regions around interest points selected; thirdly, it picks out distinctive features from the original set of interest points. The method has been implemented within a SLAM framework designed for use in a texture-rich environment such as the Great Barrier Reef. The results have shown that this method has significant advantages over other widely used methods in this specific environment. The speed of implementation is faster and the number of features needed to process is reduced.