Monocular robot navigation using invariant natural features

In this paper, we present an approach for monocular robot navigation based on natural landmarks. The scale invariant feature transform (SIFT) features are used to structure landmarks because they are invariant to image scale, rotation and translation. During learning phase, the algorithm selects the most visually salient natural landmarks in work environment on certain position. These natural landmarks are described by SITF features and save to database. When given a scene of environment during robot navigation phase, the sift-based landmark recognition algorithm is used to find corresponsive object amongst the database. If the corresponsive landmark found, they are tracked by Kanade-Lucas-Tomasi (KLT) tracker over times in order to get the relative pose information between robot and landmarks. To get more accurate relative pose information, least squares matching (LSM) method is used to get the sub-pixel matching result. The experiments results show that this approach can obtain quite accurate relative position information about natural landmark for robot navigation.