Development of an Online Algorithm for Classifying Eye-Tracking Data of Helicopter Pilots

Kurzfassung

This thesis demonstrates the feasibility of online eye movement event detection tailored for helicopter pilots. Four algorithms (I-VT, I-DT, I-KF, and I-HMM) were optimized using a synthetic dataset from a fixation-saccade invocation task study, employing a novel pipeline combining event matching with a penalization strategy to optimize algorithm parameters. The I-VT and I-KF proved most effective, outperforming Tobii’s VT filter, and were adapted for online use. Verification in the AVES flight simulator under varied visual and workload conditions confirmed reliable, interpretable classification with tolerable latency. Limitations in generalizability and configurability indicate areas for future refinement, including head-movement correction and improved usability