近年來由於嬰兒於睡眠中猝死之案例頻傳，且越小的嬰兒平均一日睡眠時間又越長，所以相關的預防顯得格外重要。家長和照護者若希望能將猝死可能性降至最低，最好的辦法是能夠時時刻刻緊盯著嬰兒，然而此建議於現實生活中明顯不可行。因此本研究以「以呼吸頻率為基礎之嬰兒意外監控系統」為基礎，改良其缺點，使其正確率更高且適用性更廣，提出並命名為「視覺式嬰兒呼吸監測系統」。將攝影機架設於嬰兒床之上方，以不會干擾到嬰兒睡眠之非接觸式監測方式，計算嬰兒呼吸頻率，並由呼吸頻率判斷嬰兒當下之呼吸是否有過快、過慢或停止呼吸之狀況，有則提出警報提醒照護者。
本論文主要提出3個改良處：自適應參數調整、放大影像相減差異和呼吸點過濾。自適應參數調整之目的在於，不同影片適用之參數不同，若無論哪支影片全部皆給予相同定值，則正確率將無法提高，因此本研究以自適應參數調整之方法，降低原先系統因驗證問題所導致之錯誤率。放大影像相減差異之目的在於，拍攝影片於畫面平均亮度低之情況下，兩兩連續影像相減結果將遠不如平均亮度正常之影片明顯，而此將造成所偵測出之POI(取Point of interest縮寫，即所偵測出可用於量測呼吸之點)無法真正凸顯影像強度值變化之部分，因此本研究提出以放大影像相減差異之方法，放大兩兩連續影像相減結果，再以此結果進行後續步驟。而呼吸點過濾之目的在於，通過驗證所蒐集的POI，當中仍可能因為各種因素(例如風吹干擾)而含有不具呼吸資訊之不良POI，若不將之濾除則呼吸頻率偵測正確率將受到影響，因此本研究提出以傅立葉轉換和sensitive function完成此步驟之檢查及過濾之動作。
本研究實驗結果使用兩個資料庫進行實驗，分別為嬰兒睡眠資料庫和特殊狀況嬰兒睡眠資料庫。其中嬰兒睡眠資料庫共有66段影片，採取三種不同方式分類，合計影片長度為4時13分43秒。系統尚未增加自適應參數調整機制時，呼吸頻率量測正確率為0.73，系統增加自適應參數調整機制時，呼吸頻率量測正確率為0.88，系統再加入放大影像相減差異機制後，亮度低之14支影片呼吸頻率正確率由0.64提升至0.79。經過三處改良後，亮度正常之52支影片呼吸頻率正確率由0.75提升至0.90，亮度低影片之正確率由0.64提升至0.71，總影片數呼吸頻率正確率由0.73提升至0.88。而特殊狀況嬰兒睡眠資料庫則有100支影片，全數於關燈狀態下拍攝(然而若於白天拍攝且窗外有光線透入，則仍可能是為亮度值為正常)，採取兩種分類方式(亮度和風吹與否)，合計影片長度2時37分21秒。100支影片呼吸頻率正確率由0.44提升至0.77，其中亮度正常且有風吹之11支影片呼吸頻率正確率由0.82提升至1，亮度正常而無風吹之10支影片呼吸頻率正確率由0.8提升至1；亮度低且有風吹之20支影片呼吸頻率正確率由0.45提升至0.8，亮度低而無風吹之59支影片呼吸頻率正確率由0.31提升至0.68。
由於呼吸方面之異常處，除了頻率快、慢和停止外，尚有呼吸過度、庫斯模式呼吸、陳潮氏呼吸和畢歐式呼吸等異常呼吸方式。因此本研究未來希望能進一步找出頻率以外之特徵，使系統除了具備判斷呼吸頻率是否過快、過慢與窒息外之安全監測功能外，亦能透過長時間記錄嬰兒呼吸模式，發現嬰兒呼吸上之疾病徵兆，以達及早治療之目的。

Because the infant death in the sleeping suddenly is often heard in recent years, and the smaller the infants the more sleep time the infants usually take in one day, it is particularly important about prevention. The best way to minimize the possibility of the infant death is that parents or caregivers stare at the baby all the time. However this opinion is impossible in the real life. Therefore, the study presents“A Vision-Based Infant Respiratory Monitoring System”, an improved respiratory monitoring system based on the research of“An Infant Monitoring System Based on a Respiratory Frequency Detection Approach” to achieve the higher accuracy and the broader applicability. The camera is set up on the top of the crib, and the system calculates the infant respiratory frequency with a non-contact monitoring way. If the frequency is too fast, too slow or breath-stopping, the system will raise alarm alerts to the caregivers.
The proposed system operates in three major improvements, including adaptive parameter adjustment, difference enlargement, and POI(Point of interest) filtering. Difference sequences need to be fitted the difference parameters. If all sequences are given the same parameters, then the accuracy will not be improved. Therefore, this study proposes an adaptive parameter adjustment to reduce the error rate due to the original system of verification problems caused. When the average brightness of the screen is low, the subtraction result of two consecutive images will be unclear, and that will result in the detection of the POI(Point of interest) can not really highlight the diversification of the image intensity values. So this study proposes the method to enlarge the differences beween two consecutive images subtraction. Because sometimes there are some bad POIs passing the verification and they interfere the respiratory frequency calculation, this study proposes the method to check and filter the bad POIs by fast Fourier transform and the sensitive function.
This study constructs two database for the experiment. There are 66 sequences with 4 hours and 13 minutes and 43 seconds in the infant sleeping sequence database. Before the system is added the adaptive parameter adjusting, the respiratory frequence measurement accuracy is 0.73. After the system increases the adaptive parameter adjusting, the respiratory frequence measurement accuracy is 0.88. Then the system is added the difference enlargement, the accuracy of 14 low brightness sequences is improved from 0.64 to 0.79. After all three improvements are added, the accuracy of 52 normal brightness sequences is improved from 0.75 to 0.9, the accuracy of 14 low brightness sequences is improved from 0.64 to 0.71, and the accuracy of total sequences was improved from 0.73 to 0.88. There are 100 sequences with 2 hours and 37 minutes and 43 seconds in the special case infant sleeping sequence database. These sequences are totally shooting in the off-turning light (However the brightness value may still be normal due to the sun light shines into). The accuracy of total 100 sequences is improved from 0.44 to 0.77. The accuracy of 11 normal brightness with wind blowing sequences is improved from 0.82 to 0.1. The accuracy of 10 normal brightness sequences (no wind blowing) is improved from 0.8 to 0.1. The accuracy of 20 low brightness with wind blowing sequences is improved from 0.45 to 0.8. The accuracy of 59 low brightness sequences is improved from 0.31 to 0.68.
The abnormal of respiration is not only shown with the too fast frequency, too slow frequency or breath-stopping, but also there are other kinds of breathing problems like hyperventilation, Kussmaul's breathing, Cheyne - Stokes respiration, and Biot's breathing. Therefore, this study hopes for the future to further identify the characteristics other than the frequency, the system is with the safety monitoring functions to determine whether the respiratory rate is too fast, too slow or suffocation, through a long record of infant breathing patterns, the system can find the signs of diseases from the baby breath in order to achieve the purpose of early treatment.

[Cao07] H. Cao, L. C. Hsu, T. Ativanichayaphong, J. Sin, and J. C. Chiao, “A Non-Invasive and Remote Infant Monitoring System Using CO2 Sensors,” IEEE Sensors Journa, pp. 989–992l, October 2007.
[Gu12] C. Gu, R. Li, H. Zhang, A. Y. C. Fung, C. Torres, S. B Jiang, and C. Li, “Accurate Respiration Measurement Using DC-Coupled Continuous-Wave Radar Sensor for Motion-Adaptive Cancer Radiotherapy,” IEEE Transactions on Biomedical Engineering, vol. 59, no. 11, pp. 3117–3123, November 2012.
[Kol13] B. L. Koley, and D. Dey, “Real-Time Adaptive Apnea and Hypopnea Event Detection Methodology for Portable Sleep Apnea Monitoring Devices,” IEEE Transactions on Biomedical Engineering, vol. 60, no. 12, pp. 3354–3363, December 2013.
[Kuo10] Y. M. Kuo, J-S. Lee, and P. C. Chung, “A Visual Context-Awareness-Based Sleeping-Respiration Measurement System,” IEEE Transactions on Information Technology in Biomedicine, vol. 14, no. 2, pp. 255–265, March 2010.
[Mer09] C. R. Merritt, H. T. Nagle, and E. Grant, “Textile-Based-Capacitive Sensors for Respiration Monitoring,” IEEE Sensors Journal, vol. 9, no. 1, pp. 71–78, January 2009.
[Min10] S. D. Min, J. K. Kim, H. S. Shin, Y. H Yun, C. K. Lee, and M. Lee, “Noncontact Respiration Rate Measurement System Using an Ultrasonic Proximity Sensor,” IEEE Sensors Journal, vol. 10, no. 11, pp. 1732–1739, November 2010.
[Rob12] C. A. Robles-Rubio, K. A. Brown, and R. E. Kearney, “Detection of Breathing Segments in Respiratory Signals,” Proceedings of the 34th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, San Diego, California, USA, pp. 6333–6336, September 2012.
[Sha14] D. Shao, Y. Yang, C. Liu, F. Tsow, H. Yu, and N. Tao, “Noncontact Monitoring Breathing Pattern, Exhalation Flow Rate and Pulse Transit Time,” IEEE Transactions on Biomedical Engineering, vol. 61, no. 11, pp. 2760–2767, November 2014.
[Shi10] J. H. Shin, Y. J. Chee, D. Jeong, and K. S. Park, “Nonconstrained Sleep Monitoring System and Algorithms Using Air-Mattress with Balancing Tube Method,” IEEE Transactions on Information Technology in Biomedicine, vol. 14, no. 1, pp. 147–156, January 2010.
[Wan05] Y. Wang, H. Baghaei, H. Li, Y. Liu, T. Xing, J. Uribe, R. Ramirez, S. Xie, S. Kim, and W. Wong, “A Simple Respiration Gating Technique and Its Application in High-Resolution PET Camera,” IEEE Transactions on Nuclear Science, vol. 52, no. 1, February 2005, pp. 125–129.
[1] 行政院衛生署中央健康保險局，「推動及應用國際疾病分類第十版ICD-10-CM/PCS於臨床疾病分類計畫」之ICD-10-CM/PCS疾病分類編碼指引，2011年。
[2] 倪麗娟，溫馨守護伴兒行-降低嬰兒猝死症候群風險照護措施，幼教新知電子報第六十七期，2012年。
[3] 衛生福利部統計處，死因統計(97-102年度)，
http://www.mohw.gov.tw/cht/DOS/Statistic.aspx?f_list_no=312&fod_list_no=5150，2013年。
[4] 謝欣紘，An Infant Monitoring System Based on a Respiratory Frequency Detection Approach以呼吸頻率為基礎之嬰兒意外監控系統，國立臺灣師範大學碩士論文，2011年。
[5] 財團法人藥品查驗中心，醫療器械分類規則(局令第15號)， 2000年。
[6] 簡易型UBabyCare聰明嬰兒床墊，http://ubabycare.com/product2.htm。
[7] 蔡孟翰、游禮碩，The Study of Obstacle detection with Supersonic Sensors超音波障礙物之偵測，逢甲大學自動控制工程學系專題製作專題論文，2009年。
[8] 張芳青、黃秀園，Research Analysis Wireless Communication's Environment to Human Health Affect無線通訊環境對人體健康影響的應用分析，南台灣2005健康照護學術研討會，2005年。
[9] 田金、許鋒，The application of medical equipment’s period maintenance and quality control in operation department醫療設備的周期巡檢和質量控制在潔淨手術部中的應用，2009年。
[10] 鄭智銘，BIOMAT睡眠多重生理監測墊設計製作，元智大學老人福祉科技研究中心，2004年。
[11] 黃勇、張玉謙、劉健民、傅佳萍、宋亮、簡越，一種用于無創呼吸機生產線自動化檢測的設備，http://www.xtexu.com/CN05/201220493171.html，2013年。
[12] Pulse Plus心跳偵測運動錶產品規格，http://www.pcstore.com.tw/unilife/M00095315.htm。
[13] 阮致維，P-Series PS2睡眠多項生理檢查儀介紹，
http://designer.mech.yzu.edu.tw/articlesystem/article/compressedfile/(2005-04-03)%20P-Series%20PS2%E7%9D% A1%E7%9C%A0%E5%A4%9A%
E9%A0%85%E7%94%9F%E7%90%86%E6%AA%A2%E6%9F%A5%E5%84%80%E4%BB%8B%E7%B4%B9.aspx?ArchID=747，2005年。
[14] 呼吸評估，國家網路醫院，http://hospital.kingnet.com.tw/service.php?mob=modload&name=diyonline&file=3-2。