Indoor localization systems have attracted considerable attention recently. A lot of works have used wireless signals from existing base stations to track users' locations. The major challenge to such systems is the signal-drifting problem. A promising direction to conquer this problem is to fuse the tracked wireless signals with inertial sensing data. In this work, we consider location tracking in a multi-floor building, which we call a 2.5-D space, by taking wireless signals, inertial sensing data, and indoor floor plans of a 2.5-D space as inputs and building a SPF (sensor-assisted particle filter) model to fuse these data. Inertial sensors are to capture human mobility, while particles reflect our belief of the user's potential locations. Our work makes the following contributions. First, we propose a model to partition a 2.5-D space into multiple floors connected by stairs and elevators and further partition each floor, according to its floor plan, into logical units connected by passages. Second, based on the 2.5D space model, we then propose particle sampling and resampling mechanisms over the logical units using wireless signal strengths and inertial sensing data to adjust our beliefs of the user's potential locations. Third, to conquer the signal-drifting problem, we propose a weighting mechanism to control the distribution of particles based on user's activities of walking on grounds/stairs and taking elevators. A prototype has been developed and tested to verify the model and its accuracy.