Localisation using an image-map

摘要

This paper looks at the problem of building a visual map for later localisation using images from a typical digital camera. The main objective is an ability to query and match images in general position against a large image data set or an image-map. We have achieved a quick localisation in terms of finding a label of the map by finding similar images in the data set. We use affinely invariant Harris corners and sift descriptors to represent 2d images. A kd-tree is used for indexing, matching, and grouping the data set to form a visual map database. For this application, we have improved a voting technique for a 3d structure and a run-time efficiency of kd-tree to allow a quick finding of similar images and a quick localisation. The result can be applied to a more generic localisation (for mobile robotics) and may be integrated in a visual odometry system.