PhD Forum: Sensor Fusion for Outdoors Augmented Reality on Android

Augmented Reality on smartphones in outdoors environments is a challenging task. Traditional SLAM methods based on local feature tracking are severely hindered by the drastic illumination changes, appearance variations and repeating textured patterns (e.g. leaves and grass) typical of the outdoors setting. We explore a sensor fusion approach that exploits visual cues and the smartphone's inertial sensors. Specifically we use mountain profiles based registration, which is computationally complex but very robust to appearance variations and local features tracking, which is fast but less reliable. A first iteration of the algorithm is being used as the core of an AR application for augmenting panoramic views of cities with historical information.