( 您好!臺灣時間:2021/03/04 09:05
字體大小: 字級放大   字級縮小   預設字形  


研究生(外文):Jhe-Fu Lin
論文名稱(外文):A Real Time Face Occlusion Detection Surveillance Assistant System for Automated Teller Machine using Deep Learning
指導教授(外文):Chuin-Mu WangGeng-Cheng Lin
外文關鍵詞:Deep LearningConvolutional Neuron NetworksObject DetectionFace Occlusion
  • 被引用被引用:0
  • 點閱點閱:207
  • 評分評分:系統版面圖檔系統版面圖檔系統版面圖檔系統版面圖檔系統版面圖檔
  • 下載下載:0
  • 收藏至我的研究室書目清單書目收藏:0
近年來ATM 犯案事件頻傳,一般的犯罪者在做案時,為了避免暴露身分會採取遮蔽臉部的行為,而目前的防範措施是直接在自動提款機置入攝影機,當發生不當領取事件發生後,才調閱監視錄影紀錄來查案,不過由於大部分自動提款系統無論臉部是否遮蔽皆可進行提領動作,要能夠識別歹徒面容並能夠有效率的完成逮捕行動顯得十分不易。
本篇論文借鑒YOLOv2[14]這篇方法來處理物件偵測 (object detection) 問題並加以改良,主要針對具有穿戴帽子、太陽眼鏡、口罩、安全帽等,不同排列組合的臉部遮蔽資訊作為學習資料,我們利用經過預訓練(pre-train)初始化網路的權重來接近最優權重,再次訓練(training)我們所建立的資料庫來學習臉部遮蔽物特徵,最後利用經過改良過後的多目標預測結果過濾方法等一系列過程來完成一個具有臉部遮蔽物辨識能力的網路模型。
In recently, ATM crime has frequently occurred, most of the suspect will conceal their identity by wearing some occlusion object. However, this is allowed that the user is cover their face or not presently. To deal with this issue, the police mostly investigate the case by access the video tape captured from the security camera frame by frame after the incident was reported. Besides the security, there is no way to solve this immediately at this stage.
According to the issue, we mentioned above, we present a new point of view to deal with it. Out concept is: “If we could verify whether there is one or more person has high masking level in the visible field in the video, then we could doubt that person is probably going to crime”. Depends on this, we could notice the security to make a decision in time and take action more efficiently.
In this paper, we present a convolution-neural-network which is base on YOLOv2 for object detection, mainly focus on the occlusion images which involve cap, sunglasses, mask, helmet and their combination as our training data. Also, at training, we initialized our networks by pre-training a model, then using our own made data to train our networks again. At testing, we replace the traditional way of filtering the wrong prediction and apply some technique to improve the performance of recognition ability.
中文摘要 i
致謝 v
目錄 vi
表目錄 viii
圖目錄 ix
第一章、緒論 1
1.1 研究背景 1
1.2 論文目標 2
1.3 論文架構 3
第二章、文獻探討 4
2.1 深度學習 4
2.2 類神經網路 5
2.3 卷積神經網路 10
2.4 Region-Based CNN 14
2.5 YOLO[2] 16
2.6 SSD[6] 21
2.7 YOLOv2[14] 25
第三章、網路設計與改良 30
3.1 網路架構 30
3.2 訓練階段 32
3.3 軟式非最大抑制(Soft Non-Maximum Suppression [13]) 36
第四章、系統建置工作與流程 39
4.1 深度學習開發工具 39
4.2 硬體規格及系統流程 41
第五章、資料庫建立 42
第六章、實驗細節與成果 44
6.1 評測標準 44
6.2 實驗結果 46
第七章、結論 55
參考文獻 56
[1]K. Fukushima, “Neocognitron: A Self-organizing Neural Network Model for a Mechanism of Pattern Recognition Unaffected by Shift in Position”, in Biol. Cybern., 1980.
[2]J. Redmon, S. Divvala, R. Girshick, A. Farhadi, “You only look once: Unified, real-time object detection”, in CVPR, 2016.
[3]C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich, “Going deeper with convolutions”, in CoRR, 2014.
[4]A. Krizhevsky, I. Sutskever, and G. Hinton, “Imagenet classi-fication with deep convolutional neural networks”, in NIPS, 2012.
[5]M. Lin, Q. Chen, and S. Yan, “Network in network”, in CoRR, 2013.
[6]W. Liu, D. Anguelov, D. Erhan, C. Szegedy, and S. Reed, “SSD: single shot multibox detector”, in CoRR, 2015.
[7]K. Simonyan and A. Zisserman. “Very deep convolutional networks for large-scale image recognition”, in ICLR, 2015.
[8]M. Holschneider, R. Kronland-Martinet, J. Morlet, P. Tchamitchian, “A real-time algorithm for signal analysis with the help of the wavelet transform”, in CPT, 1998.
[9]S. Ioffe and C. Szegedy, “Batch normalization: Accelerating deep network training by reducing internal covariate shift”, in ICML, 2015.
[10]G. Hinton, N. Srivastava, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, “Improving neural networks by preventing co-adaptation of feature detectors”, in CoRR, 2012.
[11]J. Donahue, Y. Jia, O. Vinyals, J. Hoffman, N. Zhang, E. Tzeng, and T. Darrell, “Decaf: A deep convolutional activation feature for generic visual recognition”, in ICML, 2014.
[12]N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, “Dropout: A Simple Way to Prevent Neural Networks from Overfitting”, in JMLR, pp. 1929-1958, 2014.
[13]N. Bodla, B. Singh, R. Chellappa, L. Davis, “Improving Object Detection With One Line of Code”, in CVPR, 2017.
[14]J. Redmon and A. Farhadi. Yolo9000, “Better, faster, stronger”, in CVPR, 2016.
[15]K. Cho, B. Merrienboer, C. Gulcehre, F. Bougares, H. Schwenk and Y. Bengio, “Learning phrase representations using RNN encoder-decoder for statistical machine translation”, In EMNLP, 2014.
[16]G. Hinton and R. Salakhutdinov. “Reducing the dimensionality of data with neural networks”, in Science, 2006.
[17]H. Lee, R. Grosse, R. Ranganath, and A.Y. Ng, “Convolutional deep belief networks for scalable unsupervised learning of hierarchical representations”, in ACM, 2009.
[18]J. MacQueen, “Some methods for classification and analysis of multivariate observations”, in BSMSP, 1967.
[19]R. Girshick, J. Donahue, T. Darrell, and J. Malik. “Region-based convolutional networks for accurate object detection and segmentation”, in TPAMI, 2015.
[20]J. Uijlings, K. van de Sande, T. Gevers, and A. Smeulders,“Selective search for object recognition,” in IJCV, 2013.
[21]P. Felzenszwalb and D. Huttenlocher. “Efficient Graph-Based Image Segmentation”, in IJCV, 2004.
註: 此連結為研究生畢業學校所提供,不一定有電子全文可供下載,若連結有誤,請點選上方之〝勘誤回報〞功能,我們會盡快修正,謝謝!
第一頁 上一頁 下一頁 最後一頁 top
系統版面圖檔 系統版面圖檔