Indoor location estimation based on robust floor fingerprint identification
Kaoru Uchida, Satoru Fujita · 2017
This paper presents our research on indoor location estimation based on the identity of floor surface patterns, which we call “floor fingerprints,” calculated from a photographic image of the floor taken by a user. Because floor textures generally appear to lack sufficient detail concerning surface features, it may seem impossible for general feature detection algorithms to find matching pairs of features in the images of the same floor taken from different angles and under different lighting conditions. We demonstrate, however, that use of a preprocessing image filter provides sufficient detail to the photo images to allow detection of paired features of floor textures. We also show that, although filtering reveals many noisy matching pairs irrelevant to the set of valid pairs in local keypoint descriptors, it is possible to discover a valid image-to-image correspondence from such noisy data using a selection algorithm based on histogram voting. We show, through our performance evaluation using 116 reference images and 426 query images, that precise localization is possible with 99.5% accuracy using horizontally-held smartphone devices, and 97.7% accuracy even with naturally-held devices. This paper presents our proposed approach and prototype implementation, proves its feasibility based on experimental results, and discusses its applicability to real-world indoor location applications.