Automated recognition of videotaped neonatal seizures using robust motion tracking methods that adjust to illumination and contrast changes

This paper presents motion trackers developed to quantify motion in video recordings of infants monitored for seizures. The proposed formulation relies on a variety of block motion models and can be used to develop robust motion trackers that adjust to illumination and contrast changes. The resulting motion trackers are utilized to extract motion trajectory signals, which provide the basis for selecting quantitative features that convey some unique behavioral characteristics of neonatal seizures. Such quantitative features provide the basis for training feedforward neural networks to recognize neonatal seizures.