在國內根據行政院衛生署公布民國100年的十大死因中,中風為台灣第三大死因,中風發生後大多數病患會伴隨著半邊肢體感覺障礙、半邊身體偏癱等症狀,所以需要進行復健評估及長期追蹤,往往需耗費許多時間及資源。隨著時代的進步數位資料儲存以及影像擷取處理技術成熟穩健、圖像識別演算法也越來越進步,因此本研究透過影像擷取裝置取得之動作影像序列,進行復健動作辨別及評分之研究。 本研究透過影像環境的建置以及整合現行中風復健評估量表,進行動作影像資料擷取並建立動作參數資料庫;資料經預處理後,依固定時間長度取框並計算其座標平均值做為運動學參數;運用逐步迴歸及線性鑑別分析進行動作分類並建置常模資料庫;最後利用動態時間校正演算法進行動作比對分析。本研究方法之實驗結果,組內資料比對中實驗資料與測試資料的平均比對率皆高達98.86%以上,而組間資料比對中比對率可以達到89.09%,顯示所提之方法可有效做為復健動作之量化評估比對。
According to the data announced by Executive Yuan Department of Health at 2011, Stroke is the third leading cause of death in Taiwan. The rehabilitation costs a lot of resources as conduct half of the physically disabled or hemiplegic side of the body may occur and a long-term follow-up and evaluation need to be conducted. Recent improvements in image processing and data analysis enable us for the automatic identification and quantification of rehabilitation exercises. In this study, an image sensing environment was setup for capturing the actions based on the integration of conventional assessment scale for stroke rehabilitation. The image data was tagged and transformed into kinematic parameters according to the frame-based averages of the locations of 6 major joints. Stepwise regression and linear discriminant analysis was applied for feature selection and action classification; the judgment of action completeness was determined by using dynamic time warping algorithm. The experimental results for intra-group of data showed that the accuracy of training and testing movement matching achieved an average of 98.86%; the inter-group results was about 89.00%. The experimental results encouraged us that the proposed approach is capable for the quantification and identification for rehabilitation movements.