Fusion of Odometry and Visual Datas to Localization a Mobile Robot Using Extended Kalman Filter

Abstract

This paper presented a localization system for mobile robots using fusion of visual data and odometer data. The main contribution is the modeling of the optical sensor made such a way that it permits using the parameters obtained in the image processing directly to equations of the Kalman Filter without intermediate stages of calculation of position or distance. Our approach has no pretension to be general, as it requires a flat floor with lines. However, in the cases where can be used (malls, museums, hospitals, homes, airports, etc.) when compared with another approach using geometric correction was more efficient. As future works, we intend: to improve the real-time properties of the image processing algorithm, by adopting some of the less time consuming variants of the Hough transform; Replace the Kalman Filter by a Filter of Particles, having in view that the latter incorporates more easily the nonlinearities of the problem, besides leading with non-Gaussian noises; Develop this strategy of localization to a proposal of SLAM (Simultaneous Localization and Mapping), so much that robot is able of doing its localization without a previous knowledge of the map and, simultaneously, mapping the environment it navigates.