AI影像辨識技術於日間照顧中心之應用－先導型研究

背景：日照中心對於失能長者是不錯的選擇，導入智慧科技應用於社區長期照顧場域可降低照顧人力負擔。目的：本研究收集人體進行伸手取物動作的影像，目的在於確認AI影像辨識技術是否可正確分類受測者的平衡動作為正常或異常。方法：實驗收案場域主要在某日照中心內進行，共收案數為16名，其中6位被標記為異常，其餘皆被標記為正常，然後以資料探勘及視覺分析軟體進行影像資料分析、特徵值嵌入、建立分類模型及預測結果。結果：類神經網路分類器有最佳的分類準確度，準確率達0.938。表現次佳的分類器是羅吉斯回歸，準確率達0.875，隨機森林演算法的分類準確度為0.750，至於支持向量機的分類準確度為0.625，是四種分類器中最低者。結論：本先導性研究結果發現類神經網路分類準確度最高，羅吉斯回歸則次之。研究同時發現影像拍攝角度及距離遠近可能影響分類之準確率。由於僅收集16位受試者之影像，結果僅供研究者自我參考並規劃未來研究中實驗設計必須考慮之細節。

關鍵字

失能長者；伸手取物動作；類神經網路

並列摘要

Background: The day care center is a good choice for disabled elders in Taiwan. The application of smart technology to the long-term care field in the community can reduce the loading of caregivers. Purpose: This study collected images of the subject which was doing forward reach movement. The purpose is to confirm whether the AI can correctly classify the subjects' balance movements as normal or abnormal based on image recognition technology. Method: The experiment was carried out in a day care center. A total of 16 cases were received, of which 6 were marked as abnormal, and the rest were marked as normal. Then we used data mining and visual analysis software to analyze the image data, embed, build classification models and predicted results. Results: The Neural Network classifier had the best classification accuracy, with an accuracy rate of 0.938. The next best classifier was Logistic Regression with an accuracy of 0.875, the Fandom Forest algorithm had a classification accuracy of 0.750, and the Support Vector Machine had a classification accuracy of 0.625, the lowest among the four classifiers. Conclusion: The results of this pilot study found that Neural Network classification has the highest classification accuracy, followed by Logistic Regression. The study also found that the image shooting angle and distance may affect the classification accuracy. As images of only 16 subjects were collected, the results are only for the researcher's self-reference and details that must be considered in the experimental design in planning future studies.

並列關鍵字

disabled elders ； forward reach ； Neural Network

參考文獻

顏章伊（2008）。虛擬實境平衡訓練對巴金森氏症病患姿勢穩定度及雙項任務表現之療效。 https://doi.org/10.6342/NTU.2008.02422

魏大森（2008）。老年人跌倒的篩檢與評估。台灣老年醫學暨老年學雜誌，3（2），91–105。https://doi.org/10.29461/TGG.200805.0003

Aruin, A. S., Nicholas, J. J., & Latash, M. L. （1997）. Anticipatory postural adjustments during standing in below-the-knee amputees. Clinical Biomechanics, 12（1）, 52-59. https://doi.org/10.1016/S0268-0033（96）00053-8

Bleuse, S., Cassim, F., Blatt, J. L., Labyt, E., Bourriez, J. L., Derambure, P., Destée,A & Defebvre, L. (2008). Anticipatory postural adjustments associated with arm movement in Parkinson’s disease: a biomechanical analysis. Journal of Neurology, Neurosurgery & Psychiatry, 79(8), 881-887.https://doi.org/10.1136/JNNP.2006.107318

Bussey, M. D., Aldabe, D., Shemmell, J., & Jowett, T. （2020）. Anticipatory postural control differs between low back pain and pelvic girdle pain patients in the absence of visual feedback. Human Movement Science, 69, 102529. https://doi.org/10.1016/j.humov.2019.102529

主題瀏覽