以肢體動作分析為基礎之新生兒意外監控系統

Abstract

科技日益更新，近年來基於人身的安全需求，自動化即時監控系統的相關研究正在快速的成長，一套監視系統可以幫助我們監控戶外與室內的環境狀況，甚至於面對現在少子化社會的到來，新生兒居家安全的監控與追蹤更顯得重要。 新生兒常被認為是脆弱、無助且無法明確表逹自己的需求，因此比一般病人或老年人更需要仔細且長時間的照顧與看護，也較常因照顧者的微小疏忽而導致生病或意外事故，造成新生兒父母親或照顧者很大的壓力與負擔。 本研究主要應用在分析新生兒的動作，在視覺式監控系統的監控下避免新生兒發生意外。監控攝影機架設於嬰兒床的正上方，系統讀入連續時間的新生兒動作影像，偵測新生兒肢體部位並與前一個時間的偵測結果追蹤比對，並利用新生兒肢體動作示意圖(posture map)來表示追蹤比對的結果。 新生兒的動作關係圖是由連續的肢體動作示意圖所構成，可儲存於資料庫內。當異常的情況發生時，例如新生兒的肢體動作示意圖不存在動作關係圖內或是符合動作關係圖內危險的節點時，系統將會發出警告通知照顧者前來關心。This paper presents a vision-based infant-monitoring system based on infant behavior analysis approach to avoid the occurrence of the infant’s accidents. In this study the video camera is set above the crib to capture the video sequence of the infants. The system first preprocesses the input sequence to remove the noises and reduce the influence of lights and shadows. Second, the infant-monitoring system detects the infant’s limbs and matching them to the pre-defined posture maps. These posture maps can be regard as a node and be linked to construct a motion graph. The motion graph describes the current infant’s motion. All the motion graphs can be merged into a behavior graph to describe the infant’s behaviors ever occurring before. If some unusual situations happen, for example, an input limb map does not exist in the behavior graph or be matched to a danger node then the system then alerts the baby-sitter to take care of the infant immediately. Since the infants grow very fast and their growing processes are various, the behavior graph should be updated to fit the current behaviors of infants. The experimental results show that the method has the ability to work robustly in real-time.