Extract Video Features To Detect Neonatal Epilepsy In Real Time

Newborns with epilepsy do not know how to correctly express their symptoms,the only method is to observe and judge by family members and doctors.Traditional detections of neonatal epilepsy are divided into electroencephalogram(EEG)and motion detection,which are time-consuming,expensive,restrictive and invasive,and there is a risk of infection in newborns.We adopt video detection to capture the newborn's motion videos through a smartphone and then process the video signal to detect unusual movements.Video detection is accurate,fast,and low-cost,and it is non-contact,non-restrictive,and non-invasive.Moreover,newborns have no skin-related complications,and their physical movement is not limited.Video detection compensates for the defects of the traditional detections,which is the best choice to detect neonatal epilepsy.This research mainly studies clonic epilepsy,which is characterized by the periodic reciprocating movement of various parts of the body.Therefore,we can judge whether the newborn has epilepsy by detecting the periodicity of the movement of the various parts of the body.This research applies the perceptual hash method to extract the key frame sequence to remove the redundant information;the background difference method and the automatic threshold method are applied to separate the target from the background;The interframe difference method and the flat disc element are applied to remove the noise and the non-moving part;three methods are applied for periodic detection,namely,the average luminance signal extraction method,the black pixel point statistics method,the centroid coordinate extraction method,and the method of combining the black pixel point statistics and average luminance signal extraction method is selected as the periodic detection method of the thesis by contrast;Finally,the real newborn videos are applied for detection and verification.Characteristics: The perceptual hashing method is applied to extract key frames,which improves the traditional perceptual hashing method,the ideal threshold is determined and the detection efficiency is improved;Denoising and extracting moving parts apply flat disc element closure operations in morphological methods more precisely;The traditional average luminance signal extraction method of color image is improved,and the average luminance signal extraction method of gray image is proposed,which is simpler and more effective in periodic detection;The black pixel point statistics method is proposed;Finally,the three periodic detection algorithms areapplied to the real-time neonatal epilepsy detection,and the relevant curves are drawn,which is more intuitively determined whether the newborns have epilepsy.