Thanks to the rapid development of mobile sensing technologies (like GPS, GSM, RFID, accelerometer, gyroscope, sound and other sensors in smartphones), the large-scale capture of evolving positioning data (called mobility data or trajectories) generated by moving objects with embedded sensors has become easily feasible, both technically and economically. We have already entered a world full of trajectories. The state-of-the-art on trajectory, either from the moving object database area or in the statistical analysis viewpoint, has built a bunch of sophisticated techniques for trajectory data ad-hoc storage, indexing, querying and mining etc. However, most of these existing methods mainly focus on a spatio-temporal viewpoint of mobility data, which means they analyze only the geometric movement of trajectories (e.g., the raw ‹x, y, t› sequential data) without enough consideration on the high-level semantics that can better understand the underlying meaningful movement behaviors. Addressing this challenging issue for better understanding movement behaviors from the raw mobility data, this doctoral work aims at providing a high-level modeling and computing methodology for semantically abstracting the rapidly increasing mobility data. Therefore, we bring top-down semantic modeling and bottom-up data computing together and establish a new concept called "semantic trajectories" for mobility data representation and understanding. As the main novelty contribution, this thesis provides a rich, holistic, heterogeneous and application-independent methodology for computing semantic trajectories to better understand mobility data at different levels. In details, this methodology is composed of five main parts with dedicated contributions. Semantic Trajectory Modeling. By investigating trajectory modeling requirements to better understand mobility data, this thesis first designs a hybrid spatio-semantic trajectory model that represents mobility with rich data abstraction at different levels, i.e., from the low-level spatio-temporal trajectory to the intermediate-level structured trajectory, and finally to the high-level semantic trajectory. In addition, a semantic based ontological framework has also been designed and applied for querying and reasoning on trajectories. Offline Trajectory Computing. To utilize the hybrid model, the thesis complementarily designs a holistic trajectory computing platform with dedicated algorithms for reconstructing trajectories at different levels. The platform can preprocess collected mobility data (i.e., raw movement tracks like GPS feeds) in terms of data cleaning/compression etc., identify individual trajectories, and segment them into structurally meaningful trajectory episodes. Therefore, this trajectory computing platform can construct spatio-temporal trajectories and structured trajectories from the raw mobility data. Such computing platform is initially designed as an offline solution which is supposed to analyze past trajectories via a batch procedure. Trajectory Semantic Annotation. To achieve the final semantic level for better understanding mobility data, this thesis additionally designs a semantic annotation platform that can enrich trajectories with third party sources that are composed of geographic background information and application domain knowledge, to further infer more meaningful semantic trajectories. Such annotation platform is application-independent that can annotate various trajectories (e.g., mobility data of people, vehicle and animals) with heterogeneous data sources of semantic knowledge (e.g., third party sources in any kind of geometric shapes like point, line and region) that can help trajectory enrichment. Online Trajectory Computing. In addition to the offline trajectory computing for analyzing past trajectories, this thesis also contributes to dealing with ongoing trajectories in terms of real-time trajectory computing from movement data streams. The online trajectory computing platform is capable of providing real-life trajectory data cleaning, compression, and segmentation over streaming movement data. In addition, the online platform explores the functionality of online tagging to achieve fully semantic-aware trajectories and further evaluate trajectory computing in a real-time setting. Mining Trajectories from Multi-Sensors. Previously, the focus is on computing semantic trajectories using single-sensory data (i.e., GPS feeds), where most datasets are from moving objects with wearable GPS-embedded sensors (e.g., mobility data of animal, vehicle and people tracking). In addition, we explore the problem of mining people trajectories using multi-sensory feeds from smartphones (GPS, gyroscope, accelerometer etc). The research results reveal that the combination of two sensors (GPS+accelerometer) can significantly infer a complete life-cycle semantic trajectories of people's daily behaviors, both outdoor movement via GPS and indoor activities via accelerometer.