Currently, in-shoe force sensors have been widely used for step counting and gait analysis. However, it has not been realized that in-shoe force sensors are also capable of tracking walking paths. In this paper, we present ShoesLoc, an indoor walking path tracking method based on in-shoe force sensors. We show that, based on the force signals from a user's shoes, it is possible to estimate the walking direction change and the stride length of each step with machine learning techniques. We further apply a particle filter to combine this information with the constraint of barriers in floor maps, and thus can determine the walking path and the current position of the user. To solve the problem of the low accuracy caused by cumulative walking direction errors, we improve the particle filter by designing the direction correction algorithm. Moreover, we propose the weight normalization method to handle the impact of handbags and backpacks. Our experimental results show that, after a convergence phase, ShoesLoc achieves the average location error of 0.9-1.3 m. Compared with traditional indoor tracking technologies, ShoesLoc does not require the installation of wireless anchors, and has good robustness to environment changes such as the magnetic interference. CCS Concepts: • Information systems → Location based services; • Human-centered computing → Ubiquitous and mobile computing systems and tools.