Combining SURF and SIFT for Challenging Indoor Localization using a Feature Cloud

Indoor localization for smartphone users enables applications such as indoor navigation or augmented information services. Indoor localization can be achieved by using camera images to resolve the position based on a precomputed training set of images. This technique is widely known as imagebased localization. In particular, we create a feature cloud from a Structure from Motion (SfM) approach as training set. At runtime, a feature-based matching identifies similarities between a test image and the trained set in order to solve the perspective n-point (PNP) problem and compute the camera position. Since indoor environments are challenging regarding wall structure, light conditions and glass elements, we combine SIFT and SURF image features to exploit the advantages of both techniques and, thus, provide a highly robust localization technology. We can even show that our novel approach can be used for a realtime image-based localization of a smartphone using remote processing.