Reference Free Incremental Deep Learning Model Applied for Camera-Based Respiration Monitoring

The article describes a reference and training set free incrementally trained deep learning algorithm for camera-based respiration monitoring systems. uses model based discriminator to find salient areas having like periodic motion. It stores the first principle component of found waveforms into two slowly growing along with negative, uncorrelated motion patterns. Using these samples, it trains neural network classifier recognize from sudden intensive situations. had no forgetting mechanism is able adapt quickly changing patterns conditions. has been validated in total 24 hours diverse recording captured neonatal care unit (NICU) I st Dept. Pediatrics and, II. Obstetrics Gynecology, Semmelweis University, Budapest, Hungary COHFACE publicly available dataset adult subjects. clinical data evaluation resulted mean absolute error (MAE) 6.9 root squared (RMSE) 9.8 breaths per minute, respectively, MAE was below 5 minute over 50% time. assessed subjects as well estimation RMSE values 0.95 1.7 minute.