Using descriptive image features for global localization of mobile robots

Abstract

In this paper descriptive visual features based on integral invariants are proposed to solve the global localization of indoor mobile robots. These descriptive features are locally extracted by applying a set of non-linear kernel functions around a set ofinterest points in the image. To investigate the approach thoroughly, we use a set of images taken by re-assigning the robot position many times near a set of reference locations. Also, the presence of illumination variations is encountered many times inthe images. Compared to a well-known approach, our approach has better localization rate with moderate computational overhead.