Content area
In an increasingly connected world, the proliferation of positioning and tracking technologies has generated an abundance of trajectory data. This data, while valuable, presents challenges in terms of storage, processing, and knowledge extraction. This thesis addresses these challenges by exploring the field of multisensor data fusion, with a particular focus on the development of advanced algorithms for trajectory segmentation and classification. Trajectories, composed of a series of position measurements over time, often contain redundant or erroneous information due to sensor noise. Trajectory segmentation is presented as a key technique to address this problem, allowing to obtain a more concise and manageable representation of the original trajectory. This thesis performs a comprehensive review of existing segmentation algorithms, classifying them according to their search strategy (sequential, graph, window, split, fusion or combinations) and the evaluation criteria used to select key points (distance, angle, area, velocity, probability or multi-criteria). The thesis demonstrates the practical application of trajectory segmentation in two specific case studies: Trajectory-based ship classification: The impact of different segmentation algorithms on the accuracy of ship type classification is evaluated. An AIS (Automatic Identification System) trajectory dataset is used and a framework including data cleaning, segmentation, data balancing and classification is implemented. The results show that the selection of the appropriate segmentation algorithm can significantly improve the classification accuracy. Improvement of air trajectory reconstruction in the EUROCONTROL SASS-C system: Improvements are implemented in the vertical reconstruction component of the SASS-C system (Surveillance Analysis Support System for Air Traffic Control Centers). Existing limitations, such as the presence of outliers and inadequate segmentation, are addressed through the development of new outlier detection techniques and an improved segmentation algorithm. This algorithm, based on the variation of the aircraft vertical velocity, allows for more accurate identification of the different flight phases, which in turn improves the accuracy of trajectory reconstruction. This thesis argues that trajectory segmentation not only contributes to compression and efficiency, but also has great potential for the generation of contextual information. The identification of semantic segments, representing specific behaviors or events, opens new possibilities for the analysis and interpretation of trajectories. In addition, the thesis highlights the importance of machine learning techniques for real-time trajectory analysis, which would allow early anomaly detection and proactive decision making in surveillance systems.