- 學位論文
以SONFIN分類加速規所描述之身體動作
SONFIN for Physical Activities Classification with Accelerometer
摘要
跌倒是老年人最容易發生的意外在65歲以上的老人有將近5成的人曾經發生過跌倒的意外而受傷，目前三軸加速規已被廣泛利用來監測跌倒的發生。本研究不僅是要偵測跌倒狀況，還希望能監測使用者的基本的生活行動，包含有走路、坐下和跌倒。在研究中，使用三個三軸加速規分別擺放在胸、腹和大腿上，將所擷取的獨立訊號，以極座標軸產生合成的向量訊號，再以自我回歸模型(Autoregressive model)擷取特徵，做為分類的參數，以分辨出使用者的活動狀態分類器採用自我組織學習模糊類神經網路(Self-constructing Neural Fuzzy Inference Network SONFIN)，其具有網路結構和參數的自我學習調整的能力，適合於非特定性的，為比較SONFIN的性能，採用多層感知類神經網路(Multi Layer Perceptron Neural Network MLPNN)與之去比較，實驗中有10名志願者參加這一實驗。我們的研究發現，以 60階AR模型即可描述身體動作的特性，三軸加速規適合放置在人體的胸部。而SONFIN在訓練數據時得到靈敏度 (sensitivity) 為96.3 ％與特異性(specificity)為94 ％的最佳結果。在測試數據中，靈敏度為90.7 ％ 及特異性為94 ％ 的結果，都較MLPNN好。
並列摘要
The elderly are most vulnerable to a fall accident in the 65-year-old or older nearly 50 percent of the people have occurred in the fall of the accident and the injured. At present, three-axis accelerometer has been widely used to monitor the incidence of falls. This study is not only to detect a fall status, also hope to be able to monitor the user''s basic operations include walking, sit down and fall. In the study, the use of three three-axis accelerometer were placed in the chest and abdomen and thigh, will capture the signal of independence to produce synthetic Polar coordinates axis of the signal vector, and then to self-regression model (Autoregressive model) features of admission , as the classification parameters to identify a user''s activity. SONFIN to compare the performance using multilayer neural network (MLPNN) to go with them compare. 10 volunteers to participate in this experiment. Our research found that 60 to AR-order model to describe the physical characteristics of the action, three-axis accelerometer for the human body placed in the chest. SONFIN in the data received sensitivity training for 96.3% and specificity for 94% of the best results. In the test data and the results are better sensitivity and specificity of 90.7% to 94% of the results of the MLPNN than good.