Localization in urban traffic environment for mobile robots based on stereo real-time Lightweight Visual Odometry

This paper proposes the Lightweight Visual Odometry algorithm for mobile robot localization. The proposed algorithm deals with two main difficulties: real-time functionality and robustness against environmental noise. In order to obtain real-time computation we use a closed form solution that approximates the transformation between successive pairs of points. This solution is used to compute motion hypotheses for all successively detected points. In order to obtain robustness against noise, we select features from the road surface which intuitively represent features on static objects. The major key point of sustaining robustness is how to determine the road surface in variation environment (structured, non-structured, and off-road). To do this we use a previously proposed road surface detection method. The road surface detection method accurately provides the image road surface region in any environment and it allows obtaining robust inlaying features which are static. This proposed approach is very useful especially to mobile robot navigation in crowded urban traffic environments.