Optimization of Two-Phase Sampling Designs With Application to Naturalistic Driving Studies

Naturalistic driving studies (NDS) generate tremendous amounts of traffic data and constitute an important component modern safety research. However, analysis the entire NDS database is rarely feasible, as it often requires expensive time-consuming annotations video sequences. We describe how automatic measurements, readily available in database, may be utilized for selection time segments annotation that are most informative with regards to detection potential associations between behavior a consecutive critical event. The methodology illustrated evaluated on from large naturalistic study, showing use optimized instance reduce number need annotated by much 50%, compared simple random sampling.