A Low Complexity Monocular Image Matching Scheme for Efficient Indoor Positioning

Abstract

An efficient indoor positioning system (IPS) with no hardware deployment costs is presented. Anchor points in the target field are first selected as positioning references. For each anchor point and a wraparound image is captured as a landmark image, which is further divided into overlapped full-HD sized views with an angular deviation of 7.3 degrees. By using just one image obtained from a monocular camera, i.e., the positioning image, the IPS matches it with a landmark data base to determine the approximate location and the facing direction of the camera. Color calibration is performed to alleviate the influence of ambient lighting on the images. The image matching is conducted in two phases. A histogram similarity based landmark image screening process is employed to reduce matching efforts by an average of 91%. Feature points are next extracted for the second stage matching. The feature points determined by using the KAZE scheme is scaling and rotation invariant, which can effectively compensate the difference between the positioning image and the landmark image of the nearest anchor point. The closest view of the landmark image also indicates the facing direction of the camera. The experiments conducted on a campus shopping store with 27 anchor points show that the matching is almost error free if the location is within a 60cm vicinity of any anchor point.