臺灣博碩士論文加值系統

本篇論文針對偵測家中獨居老人跌倒事件，提出了一個非監督式學習的自編碼器的架構。解決因為沒有人力看照家中獨居老人，而發生跌倒事件無人知曉的問題。主要是透過家中的監視攝影機得到家中長輩活動的影像，輸入本論文所提出的非監督式自編碼器模型，即可辨認當前的活動屬於日常活動或是跌倒事件。在本論文所設計的實驗中，最佳的狀況下可以完美的區分出跌倒事件與日常事件，與其他類似的架構中，也是以本論文的結果最為突出，因此本論文對於跌倒偵測這個題目上，提出了一個有效的解決方法。

Elderly living alone without family care which usually cause some accident. In this thesis we focus on detect fall event. We propose an unsupervised Auto-Encoder model for this task. Mainly through the home surveillance cameras to get images of the elder’s home activity. Then our falling detection model will analysis the video clip to detect the current situation belong to normal activity or fall event. In the experiment, the fall event and daily event can be perfectly distinguished under the best condition. We compare with other similar architectures in same experiment, our model also gets the best performance. Therefore, we propose a solution for the falling detection task.

誌謝 i
中文摘要 ii
ABSTRACT iii
CONTENTS iv
LIST OF FIGURES vii
LIST OF TABLES x
第1章 緒論 1
1.1 研究動機及目的 1
1.2 本論文貢獻 1
1.3 文獻回顧 1
1.4 章節概要 3
第2章 背景知識簡述 4
2.1 類神經網路 4
2.1.1 卷積神經網路 6
2.1.2 反卷積神經網路 9
2.1.3 卷積自編碼器 10
2.2 異常偵測 12
第3章 跌倒狀況偵測系統架構 15
3.1 系統架構 15
3.2 跌倒偵測模型架構 16
3.3 異常狀況的偵測方式 16
第4章 模型細節與實驗設計 19
4.1 模型建立細節 19
4.1.1 汲取空間資訊CNN 20
4.1.2 汲取時間與空間資訊CAE 21
4.1.3 解碼空間資訊的DeCNN 23
4.2 訓練細節 24
4.3 訓練資料與測試資料準備 24
4.3.1 訓練資料 26
4.3.2 測試資料 27
4.4 實驗設計 27
第5章 數值結果與討論 29
5.1 訓練結果討論 29
5.1.1 模型訓練成果 29
5.1.2 三種情境比較 30
5.1.3 跌倒偵測結果分析 33
5.2 與其他方式比較 36
5.3 多參數調校 40
5.3.1 輸入模型幀數差異 40
5.3.2 Stride、MaxPooling、AveragePooling比較 41
第6章 結論 42
參考文獻 43
附錄 45

[1]S. S.Khan andJ.Hoey, “Review of Fall Detection Techniques: A Data Availability Perspective,” pp. 0–1, 2016.
[2]X.Yu, “Approaches and principles of fall detection for elderly and patient,” 2008 10th IEEE Intl. Conf. e-Health Networking, Appl. Serv. Heal. 2008, pp. 42–47, 2008.
[3]A.Yajai andS.Rasmequan, “Adaptive directional bounding box from RGB-D information for improving fall detection,” J. Vis. Commun. Image Represent., vol. 49, pp. 257–273, 2017.
[4]S.Zhou, W.Shen, D.Zeng, M.Fang, Y.Wei, andZ.Zhang, “Spatial-temporal convolutional neural networks for anomaly detection and localization in crowded scenes,” Signal Process. Image Commun., vol. 47, pp. 358–368, 2016.
[5]D.E. Rumelhart, G.E. Hinton, andR.J. Williams, “Learning Internal Representation by Error Propagation,” Parallel Distributed Processing: Explorations in the Microstructure of Cognition, vol. Vol. 1. 1986.
[6]M.Hasan, J.Choi, J.Neumann, A. K.Roy-Chowdhury, andL. S.Davis, “Learning Temporal Regularity in Video Sequences,” 2016.
[7]Y.LeCun et al., “Backpropagation Applied to Handwritten Zip Code Recognition,” Neural Comput., vol. 1, no. 4, pp. 541–551, 1989.
[8]A.Krizhevsky, I.Sutskever, andG. E.Hinton, “ImageNet Classification with Deep Convolutional Neural Networks,” Adv. Neural Inf. Process. Syst., pp. 1–9, 2012.
[9]M. D.Zeiler, D.Krishnan, G. W.Taylor, andR.Fergus, “Deconvolutional networks,” Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit., pp. 2528–2535, 2010.
[10]H.Noh, S.Hong, andB.Han, “Learning deconvolution network for semantic segmentation,” Proc. IEEE Int. Conf. Comput. Vis., vol. 2015 International Conference on Computer Vision, ICCV 2015, pp. 1520–1528, 2015.
[11]M. A. F.Pimentel, D. A.Clifton, L.Clifton, andL.Tarassenko, “A review of novelty detection,” Signal Processing, vol. 99, pp. 215–249, 2014.
[12]Y. S.Chong andY. H.Tay, “Abnormal Event Detection in Videos Using Spatiotemporal Autoencoder,” in Advances in Neural Networks - ISNN 2017, 2017, pp. 189–196.
[13]P.Ramachandran, B.Zoph, andQ.V.Le, “Searching for Activation Functions,” pp. 1–13, 2017.
[14]M.Kepski andB.Kwolek, “Fall Detection on Embedded Platform Using Kinect and Wireless Accelerometer,” pp. 407–414, 2012.