- 學位論文
基於Android行動平台之口袋型跌倒偵測系統設計與實現
The Design and Implementation of a Pocket-based Fall Accident Detector on Android Mobile Platforms
摘要
本論文提出了一套基於Android智慧型行動裝置的口袋型可攜式跌倒偵測系統。當跌倒意外發生時，系統會自動發出閃光與聲響，並通知緊急連絡人使傷患能即時獲得援助。相較於相關研究需將偵測裝置以特定方式配戴於使用者的特定部位，本系統則僅需將偵測裝置隨意地置入口袋中，大幅提升了使用上的舒適性與便利性。此外本系統採用Android手機作為演算法和系統的發展平台，系統所內建的三軸加速計與電子羅盤便可擷取使用者的活動資訊，進行人體訊號判讀。當跌倒意外事件發生的當下則可藉由裝置所內建的GPS進行定位，再透過3G網路將經緯度座標傳遞給救援中心。此外考量行動裝置運算資源相對有限，跌倒偵測演算法除須兼顧識別率外，還需考量運算複雜度。為此，我們提出了一套基於有限狀態機結合支持向量機的跌倒偵測演算法，藉由有限狀態機透過狀態機逐級檢驗各項特徵的設計便可大幅降低行動裝置的運算負擔。本論文所提出的演算法除具備極低的運算複雜度，亦能有效地克服系統置放於口袋中所造成的雜訊干擾並能準確地判別跌倒事件的發生與否。在靈敏度以及特異性指標兩方面則分別高達了96%以及99.71%，足見本系統不但具備實用性與便利性，更具備高度的準確性。
並列摘要
We propose in this dissertation a portable pocket fall accident detection system on Android-based mobile devices. When a fall accident event is detected, the system will automatically generate a flash light as well as a sound continuously, and the emergency center will be notified so that injuries can get assistance immediately. Unlike the other existing fall accident detectors that have to be worn or fastened on the user's body in a particular way, we just put the mobile device in the pocket, resulting in a better convenience and feasible for practical usage. With the built-in tri-axial accelerometer and electronic compass in the Android-based mobile device, the information about the user's activity can be easily retrieved, and then analyzed by the proposed algorithm. When a fall accidents event is detected, the user’s current position acquired by the global positioning system (GPS) will be sent to the rescue center via the 3G communication system so that the user can get medical help immediately. Considering the limited computing resources in a mobile device, we propose in this dissertation an algorithm by using a finite state machine cascaded with a support vector machines for the detection of a fall accident event. Based on the concept of a finite sate machine, the features acquired in the proposed system will be examined in a sequential manner. Once the corresponding feature is verified by the current state, it can proceed to next stage; otherwise, the system will reset to the initial state and waiting for the appearance of another feature sequence. With the proposed approach, the computational burden can be alleviated significantly. Moreover, as we will see in the experiment that the interference caused by putting the device in the pocket can be successfully conquered and a distinguished fall accident detection accuracy up to 96% on the sensitivity and 99.71% on the specificity can be obtained when a set of 400 test actions in eight different kinds of activities are estimated by using the proposed approach which justifies the superiority of the proposed algorithm.