A Scalable Matching Mechanism for Online Heterogeneous Positioning Fusion System

Abstract

Wi-Fi fingerprint positioning has the advantages of being infrastructure-less and easily accessible, but the weaknesses in terms of lower accuracy and limited Wi-Fi scanning speed are also hard to tackle. On the other hand, with the progress in computer vision and deep learning, vision-based positioning based on commonly available surveillance cameras becomes a promising solution for providing location-based services. But the major difficulty lies in checking the identity of detected people just by the captured images. This paper proposed a novel Matching Mechanism to address the identity matching problem, which associates the non-identifiable positioning sources like vision to the easily identifiable positioning sources like smart phone’s Wi-Fi. Practicalities like scalability and online operation are considered in both the design and implementation of the mechanism. As a result, the experiment not only proved the effectiveness of matching the vision-based and Wi-Fi positioning results but also showed an improvement in positioning accuracy by over 60%.