跌倒是發生危險的重要依據，不管是因為自己行走不慎發生的跌倒，還是身體不適導致的跌倒，如果沒有及時發現很可能會錯失搶救的黃金時間。 本論文欲實現在深夜光線不足環境下的安全監控，可能是室內的老人看護或是夜間的斑馬線上。實驗著重於行人跌倒的部分，若能藉由本系統即時偵測出是否有人跌倒不起，可適時增加紅綠燈的秒數，甚至是啟動警報系統通知保全或是鄰近醫院給予救援，如此一來將可大大提高搶救的成功率。 有別於一般安全監控的系統，在夜間環境下會因光線不足而增加大量雜訊，因此無法使用單純的監視錄影機去實現本系統。市面上之紅外線監視器是用紅外線輔助光源增加夜間影像擷取強度，但若在完全漆黑的環境下，取得的影像依然不夠清晰，增加影像處理之困難。 若使用熱像儀作為監視器，即使在沒有任何光源的環境下，也可以擷取出完整的人體輪廓，且可依溫度高低輕易地去除背景，利於實踐人體相關之影像處理。經實驗結果證明我們的系統在夜間斑馬線上是可行且具有良好偵測率。Falling down is an important factor for dangerous evaluation. No matter it is caused due to physical discomfort or accident. If it is not detected at real time, it will probably miss the prime time of rescuing.In this thesis, we focus on safety monitoring in the night of low-light environments. The situations occur frequently for indoor elderly care at night or pedestrian zebra crossing at night. Our work is designed mainly on falling down detection of pedestrians. If the system detects the situation that someone falls down in zebra crossing and can not climb up in a certain period of time, it will be synchronized with the traffic light controller to increase the green light period on the zebra crossing side. In the meantime, an alert is triggered. Alert is issued in the form of sound or flashing alarm whenever falling down situation is detected to warn the drivers.Different from general traffic safety system capturing images by CCD cameras, noises and environmental effects usually generate images with very poor quality and lose a lot of originally visible information at night. Hence, traditional camera-based systems do not work well. To alleviate this problem, some security systems add infrared ray to assist image capturing. However, the images are still not clear enough to assure the normal operation of most security systems in the fully dark environment. In our work, thermal cameras are adopted instead to replace traditional cameras to overcome the aforementioned problems. Thermal cameras can extract clear and complete body contours even in the absence of lights. Experimental results demonstrate the feasibility and validity of our proposed method in detecting the occurrence of falling down, especially in nighttime.
