A Technique of Natural Visual Landmarks Detection and Description for Mobile Robot Cognitive Navigation

Abstract

A necessary condition for the creation of robotic systems that would be able to replace a human in hazardous conditions is developing means of intelligent control and autonomous navigation providing the possibility of making independent decisions.The article describes a technique of natural visual landmarks detection and description for mobile robot cognitive navigation.The proposed landmarks are built by simultaneous filtering and clustering of a saliency map and a point map that can be achieved by a visual SLAM method like PTAM.As a result the natural landmarks are selected from the scene and described with point clouds and image patches from different point of views.The method significantly improves robot localization robustness (15% localization error from PTAM is completely removed) and can be used for semantic description of objects in a robot navigation map.