Map matching has been used to reduce the noisiness of the location estimates by aligning them to the road network on a digital map. A growing number of applications, e.g. energy efficient localization and cellular provider side localization, depend on the availability of only sparse and coarse-grained positioning data; leading to a challenging map matching process. In this paper, we present semMatch: a system that can provide accurate HMM-based map matching for challenging positioning traces. semMatch leverages the smartphone's inertial sensors to detect different road semantics; such as speed bumps, tunnels, and turns; and uses them in a mathematically-principled way as hints to overcome the sparse, noisy, and coarse-grained input positioning data, improving the HMM map matching accuracy and efficiency. To do that, semMatch applies a series of preprocessing modules to handle the noisy locations. The filtered location data is then processed by the core of semMatch system using a novel incremental HMM algorithm that combines a semantics-enriched digital map and the car's ambient road semantics in its estimation process. We have evaluated semMatch using traces collected from different cities covering more than 150km under different harsh scenarios including coarse-grained cellular-based positioning data, sparse GPS traces with extremely low sampling rate, and noisy traces with a large number of back-and-force transitions. The results show that semMatch significantly outperforms traditional map matching algorithms under all scenarios, with an enhancement of at least 416% and 894% in precision and recall respectively in the most difficult cases. This highlights its promise as a next generation map matching algorithm for challenging environments.
