以呼吸頻率為基礎之嬰兒意外監控系統

Abstract

近年來由於生育率的明顯降低，政府除應積極的鼓勵生育外，另一方面，新生兒意外事故的預防亦顯得格外重要。由於嬰兒的呼吸異常不容易察覺，但是卻帶給嬰兒的生長發育極大傷害，尤其是新生兒及嬰兒的睡眠呼吸暫停，極有可能是引起猝死的重要原因。因此本論文將開發一套「以呼吸頻率為基礎之嬰兒意外監控系統」，該系統把攝影機架設在嬰兒床上方利用視覺式的方式來監控嬰兒可能發生猝死的徵兆，例如呼吸是否不規律、是否暫停、是否急促等，並依各種情形發出不同程度的警示訊息即時告知照顧者，提醒照顧者檢視嬰兒的現況，希望減少意外事故的發生。 本論文主要分成四個部分：motion偵測、呼吸候選點偵測、呼吸點選取、呼吸頻率計算。本系統讀入連續嬰兒影像後，先判斷是否適合進行呼吸偵測，若影像中motion量大時表示嬰兒具有活動力；反之，motion量小時表示嬰兒於於畫面中並無明顯動靜而發生窒息的可能性較大。一旦偵測出motion量小的情況系統會從累積的影像差值中偵測呼吸候選點(candidate points)，之後將候選點偵測出的特徵進行模糊整合(fuzzy integral)篩選出合適的呼吸點。系統將呼吸點所在位置其灰階值的變化利用高斯連續模糊整合(Gaussian successive fuzzy integral)來計算影像中嬰兒的呼吸頻率。 實驗結果共有22段影片，合計影片長度約為105分鐘，其中共偵測10個不同的嬰兒。實驗結果以影像張數來計算正確率為73.37%；以整段影片來計算正確率為100%。由此可知，本系統具有效性及穩定性且能即時計算出結果警示照顧者。 本實驗可用來做呼吸的長期偵測，以期建構出一套嬰兒成長速度與呼吸頻率相關性的常模。未來可用來觀察監控中之嬰兒其發展是否符合一般常態，並可長期追蹤先天呼吸系統不健全的嬰兒其病情是否穩定等，提供治療時的必要資訊。

Sudden infant death syndrome (SIDS) is a primary cause of death for infants aged one week to one year. The SIDS rate has been declined due to the awareness of caregivers and parents but that rate is still high even in developed countries such as U.S. because of the difficulty in rescuing infant immediately. Respiration, which can reflect various signs about physiological conditions, is a basic but vital function for infants. Therefore, the study presents a monitoring system with a video camera positioned in front of an infant to non-invasively monitor respiratory frequency of that infant. The proposed system can 24 hours continuously monitor the infant to detect the unusual event occurrence of the infant’s respiration to alert the caregivers to attend the infant immediately to reduce the potential injury of SIDS. The proposed system operates in four major stages, including motion detection, candidate point extraction, respiration point selection, and respiratory frequency calculation. During motion detection stage, the system captures images from video and decides whether to conduct the following stages or not. If no obvious motion has been detected in the images, which means the infant may have the occurrence possibility of SIDS, the system will extract candidate points by spatial characteristic. Based on the extracted candidate points, the system then selects respiration points by a fuzzy integral technique with four features. These features are entropy, period, skewness, and kurtosis according to temporal characteristic. Finally,the respiratory frequency of infant is processed by means of Gaussian successive fuzzy integral method. The experimental data are obtained from 10 different infants, and the number of sequence is 22 with total length 105 minutes. The detection accuracy of infant respiratory frequency is 100%. As a result, the proposed system is robust and efficient. Moreover, health protection can be alerted in time. Furthermore, the system can monitor long-term infant respiratory frequency to construct the frequency norms corresponding to different infant growth rates. These frequency norms can help to examine whether an infant meets general standard and to recognize the development or progression of some diseases at an early stage in the future.