Using feature scale change for robot localization along a route

An appearance-based similarity measure for localizing a robot along a route is presented. This measure assesses the likelihood that the robot lies between a pair of positions where snapshot images were captured during training. The change in the scale parameter of matched SIFT features is used to determined whether the robot lies ahead or behind each snapshot. Experimental results in two different indoor environments suggest that this similarity measure will improve localization accuracy in situations where there is a large distance between snapshot positions.