A Comparison of Video-based Methods for Neonatal Body Motion Detection

Area of Science:

• Medical Technology
• Neonatal Care
• Biomedical Engineering

Background:

• Preterm infants in NICUs require continuous vital sign monitoring.
• Body motion patterns are indicators of infant maturation and clinical events.
• Current motion monitoring is intermittent and relies on clinical observation.

Purpose of the Study:

• To assess the feasibility of simple video-based methods for infant body motion detection.
• To compare the performance of different motion detection algorithms.
• To evaluate unobtrusive, non-contact motion monitoring in the NICU.

Main Methods:

• Investigated four video-based motion detection methods: background subtraction (BS), sparse optical flow (SOF), dense optical flow (DOF), and oriented FAST and rotated BRIEF (ORB).
• Utilized two datasets: 32 hours of thermal videos and 9 hours of RGB videos from preterm infants.
• Evaluated detection performance using area under receiver operating curves (AUC) and computational efficiency via run time.

Main Results:

• In thermal imaging, background subtraction (BS) achieved the highest performance (AUCs of 0.86 and 0.93).
• In RGB imaging, sparse optical flow (SOF) demonstrated superior performance (AUCs of 0.82 and 0.91).
• All tested methods showed efficiency for integration with low-resolution thermal cameras.

Conclusions:

• Simple video-based methods are feasible for detecting infant body motion in the NICU.
• Different algorithms perform optimally depending on the camera type (thermal vs. RGB).
• This technology holds promise for continuous, non-contact monitoring of preterm infants.