Sleep Apnea Detection Based on Respiratory Movement Using Harris Corner and Recurrent Neural Networks

Sleep Apnea is a temporary breathing stop during sleep, which can be detected through video observations. The principle is to make precise and efficient head and chest segments, take consistent reference points, and recognize vertical deflection patterns in time sequences. This study used the Fast Fourier Transform for filters other than the head and chest. Then, Harris Corner detects the head and chest in every frame. Then Haar Cascade got good results as feature assistance from faces and tracking with the Kanade Lucas Tomasi Algorithm. The RNN-GRU classifies the vertical pattern of each frame into three classes: No Apnea, Obstructive Sleep Apnea, and Central Sleep Apnea. The results showed that the accuracy used in the RNN-GRU is 85.18% with the Adamax technique.