Mobile Robot Self-Localization Based on Omnidirectional Vision and Gaussian Models

We present preliminary results derived from our project on robot self localization using omni directional images. In our approach, features are generated through the computation of covariance matrices that capture important patterns that relates changes in pixel intensities. The learning models used are Mixture of Gaussians and Gaussian Discriminant Analysis. The first method is used initially to test the viability of our feature vectors, and at the same time provides useful information about a natural way of clustering the images in our traning set. Once we determined a reliable set of features, we generated the Gaussian discriminant functions. We show promising experimental results obtained with the Pioneer P3-DX robot in the hallways of the School of Mathematics at the Yucatan Autonomous University in Mexico.