Mobile robot Monte-Carlo localization using image retrieval

To improve the localization capability of the mobile robot in a dynamic environment, we propose a vision-based Monte-Carlo localization approach. We apply the image retrieval technique to compute the similarity between query images and the images stored in an image database. In order to reduce the effect of images matching cased by illumination change, relational kernel functions are used to extract the features from images. During the Monte-Carlo localization, we use the visibility area of the referenced images which have been computed off-line to update the particles’ post probability. The practical experiments illustrate that our approach is able to locate the robot accurately under the dynamic environment with change especially the illumination.