The ability to accurately track pedestrians is valuable for variant application designs. Although pedestrian tracking has been investigated excessively and owned a well-suited sensing platform, the proposed solutions are far from being mature yet. Pedestrian tracking contains step counting and stride estimation two components. Step counting already has commercial products, but the performance is still unreliable and less trustworthy in practice. Stride estimation even stays in the research stage without ready solutions released on the market. Such a non-negligible gap between long-term research investigation and technique’s actual usage exists due to a series of crucial applicability issues unsolved, including design vulnerability to interfering activities, extracting purely body’s movement from additive sensor signals, and parameter training without user’s intervention. In this paper, we deeply analyze human’s gait cycles and obtain inspiring observations to address these issues. We incorporate our techniques into existing pedestrian tracking designs and implement a prototype, PTrack, using LG smartwatch. We find PTrack effectively enhances the system applicability and achieves promising performance under very practical settings.