Air Traffic Trajectories Segmentation Based on Time-Series Sensor Data
Jose Gerardo Ortile Guerrero, Jesús Garcı́a, José Manuel · Sciyo eBooks · 2010
The automation of ATC systems is a complex issue which relies on the accuracy of its low level phases, determining the importance of their validation. That validation is faced in this work with an inherently offline processing, based on a domain transformation of the noisy measurements with three different motion models and the application of machine learning and filtering techniques, in order to obtain the final segmentation into these different models. This work has analyzed and defined in depth the uniform motion model and the algorithm’s performance according to this model. The performance analysis is not trivial, since only one of the motion models in the algorithm is presented and the results obtained are, thus, only partial. Even so, these results are encouraging, having obtained good TPR and FPR values in most trajectory types, and a final number of segments which are reasonably similar to the real ones expected. Some issues have been pointed out, such as the behaviour of measurements belonging to uniform motion models when they are close to two different non-uniform segments (a typical situation during racetrack trajectories), but the final algorithm’s results are required in order to deal with these issues properly. Future lines include the complete definition of the algorithm, including the non uniform motion models and the study of possible modifications in the domain transformation, in order to deal with the introduced difficulties, along with the validation with real trajectories.