(18.206.177.17) 您好!臺灣時間:2021/04/11 03:51
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果

詳目顯示:::

我願授權國圖
: 
twitterline
研究生:謝勻詒
研究生(外文):Hsieh, Yun-Yi
論文名稱:基於深度學習之頂照式魚眼攝影機下的人物偵測
論文名稱(外文):Deep Learning Based Human Detection Under Fisheye Camera
指導教授:王才沛
指導教授(外文):Wang, Tsai-Pei
學位類別:碩士
校院名稱:國立交通大學
系所名稱:多媒體工程研究所
學門:電算機學門
學類:軟體發展學類
論文種類:學術論文
論文出版年:2018
畢業學年度:107
語文別:中文
論文頁數:31
中文關鍵詞:人物偵測魚眼鏡頭深度學習
外文關鍵詞:People DetectionFisheye CamerasDeep Learning
相關次數:
  • 被引用被引用:0
  • 點閱點閱:332
  • 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:41
  • 收藏至我的研究室書目清單書目收藏:0
人物偵測在現今視訊媒體發達 下,在各種監控 下,在各種監控 下,在各種監控 應用上都是一項重要的工作, 應用上都是一項重要的工作, 近年來,由於深度學習迅速竄紅在基下對人物進行偵測的研究也十 近年來,由於深度學習迅速竄紅在基下對人物進行偵測的研究也十 近年來,由於深度學習迅速竄紅在基下對人物進行偵測的研究也十 近年來,由於深度學習迅速竄紅在基下對人物進行偵測的研究也十 近年來,由於深度學習迅速竄紅在基下對人物進行偵測的研究也十 分的多,但 大部分的研究都是針對投射型攝影機 ( projective camera )下的影像, 反而 在頂照式 魚眼攝影機 (top-view fisheye camera)下的研究相對寥無幾 下的研究相對寥無幾 ,本篇 論文將之設為研究目標 。
在深度學習下有許多各種不同的神經網路架構,本篇論文使用基於 Mask-RCNN神經網路架構下,對頂照式魚眼攝影機的人物進行偵測。 神經網路架構下,對頂照式魚眼攝影機的人物進行偵測。 神經網路架構下,對頂照式魚眼攝影機的人物進行偵測。 神經網路架構下,對頂照式魚眼攝影機的人物進行偵測。 由於頂照式魚 由於頂照式魚 眼攝影機下的人在圖片各個位置會有不同形變, 眼攝影機下的人在圖片各個位置會有不同形變, 為了增加準確率,將魚眼攝 為了增加準確率,將魚眼攝 為了增加準確率,將魚眼攝 影機下的人 物獨立成一類 去作訓練,而本篇論文也 作訓練,而本篇論文也 作訓練,而本篇論文也 作訓練,而本篇論文也 將圖片劃分成 中心及外圍 的部 分,成兩種 分,成兩種 分,成兩種 model各別作訓練 ,再將結果 結合起來與前者作 結合起來與前者作 比對,除此之外還 做了許多 相關的 實驗,希望藉此 提升偵測率 。
Human detection is an important task in various surveillance applications under the current development of video media. In recent years, due to the rapid learning of deep learning, there have been many studies on the detection of people based on deep learning. However, most of the research is on the image of the projective camera. On the contrary, the research under the top-view fisheye camera is relatively rare. Therefore, this paper sets it as the research target.
There are many different neural network architectures under deep learning. This paper uses the Mask-RCNN neural network architecture to detect people under the top-view fisheye camera. Since the people under the top-view fisheye camera will have different deformations at various positions in the picture, in order to increase the accuracy, the people under the fisheye camera are independently trained into one class, and this paper also divides the picture into central and periphery part, divided into two models for training, and then combine the results to compare with the former. In addition, many related experiments have been done, hoping to improve the detection rate.
摘要 ............................................................................................................................................ I
Abstract ..................................................................................................................................... II
致謝 .......................................................................................................................................... III
目錄 .......................................................................................................................................... IV
圖目錄 ...................................................................................................................................... VI
表目錄 ..................................................................................................................................... VII
第一章 緒論 ............................................................................................................................. 1
1.1 背景介紹 .................................................................................................................. 1
1.2 研究動機與方法 ...................................................................................................... 1
1.3 論文架構 .................................................................................................................. 2
第二章 文獻探討 ..................................................................................................................... 3
2.1 機器學習 .................................................................................................................... 3
2.2 深度學習 .................................................................................................................... 4
第三章 研究方法 ..................................................................................................................... 9
3.1 實驗用資料庫 ................................................................................................................ 9
3.1.1 資料來源 ............................................................................................................. 9
3.1.2 Ground Truth的標記 ........................................................................................ 11
3.2 Mask R-CNN網路架構 .................................................................................................. 12
3.3實驗方法 ....................................................................................................................... 14
3.3.1 Data Augmentation ................................................................................................ 15
3.3.2 交錯場景訓練 ....................................................................................................... 15
3.3.3 同場景訓練 ........................................................................................................... 15
3.3.4 同場景不同攝影機訓練 ....................................................................................... 16
3.3.5 加入測試背景至訓練資料 ................................................................................... 16
V
3.3.6 分區訓練不同模型 ............................................................................................... 17
第四章 研究結果 ................................................................................................................... 20
4.1實驗規劃 ....................................................................................................................... 20
4.1.1 實驗環境 ............................................................................................................... 20
4.1.2 實驗資料 ............................................................................................................... 20
4.2 偵測之準確度評分方法 ........................................................................................ 21
4.3 實驗結果 ................................................................................................................ 22
4.3.1 Add New Class ........................................................................................................ 22
4.3.2 Data augmentation ................................................................................................. 23
4.3.3 交錯場景訓練及同場景訓練比較 ....................................................................... 23
4.3.4 交錯場景與同場景不同攝影機訓練比較 ........................................................... 24
4.3.5 加入背景至訓練資料 ........................................................................................... 26
4.3.6 分區訓練模型比較 ............................................................................................... 27
4.3.7 Baseline比較 ......................................................................................................... 28
第五章 結論 ........................................................................................................................... 29
參考文獻 ................................................................................................................................. 30
[1] K. He, G. Gkioxari, P. Dollar, and R. Girshick. Mask R-CNN. arXiv preprint arXiv:1703.06870, 2017.
[2] Y. Li, H. Qi, J. Dai, X. Ji, and Y. Wei. Fully convolutional instance-aware semantic segmentation. In CVPR, 2017.
[3] S. Ren, K. He, R. Girshick, and J. Sun. Faster R-CNN: Towards real-time object detection with region proposal networks. In NIPS, 2015.
[4] R. Girshick, J. Donahue, T. Darrell, and J. Malik. Rich feature hierarchies for
accurate object detection and semantic segmentation. In CVPR, 2014.
[5] R. Girshick. Fast R-CNN. In ICCV, 2015.
[6] Redmon, J., Divvala, S., Girshick, R.B., Farhadi, A.: You only look once: unified, real-time object detection. In: CVPR 2016.
[7] A.T. Chiang and Y, Wang, Human detection in fish-eye images using HOG-based detectors over rotated windows, Proc. ICME Workshops, 2014.
[8] J. Long, E. Shelhamer, and T. Darrell. Fully convolutional networks for semantic
segmentation. In CVPR, 2015.
[9] N. Dalal and B. Triggs, Histograms of oriented gradients for human detection, Proc. CVPR, pp. 886–893, 2005.
[10] L. Meienl et al. Virtual perspective views for real-time people detection using an omnidirectional camera. In IEEE IST, pp. 312–315, 2014.
[11] V.T. Nguyen, T.B. Nguyen, and S.T. Chung. ConvNets and AGMM based Real-time Human Detection under Fisheye Camera for Embedded Surveillance. In ICTP, pp.840-845, 2016.
[12] L. Meinel, M. Findeisen, M. Heß, A. Apitzsch, and G. Hirtz, Automated real-time
surveillance for ambient assisted living using an omnidirectional camera, Proc.
31
ICCE, pp. 396-399, 2014.
[13] M. Saito, K. Kitaguchi, G. Kimura, and M. Hashimoto, Human detection from f-
isheye image by Bayesian combination of probabilistic appearance models, Proc.
SICE Annual Conference, 2011. [14] Demiröz B. E., Arı İ., Eroğlu O., Salah A. A., Akarun L., Feature-Based Tracking
On A Multi-Omnidirectional Camera Dataset, ISCCSP12, Rome, Italy, 2012.
[15] The PASCAL Visual Object Classes Homepage http://host.robots.ox.ac.uk/pascal/VOC/
連結至畢業學校之論文網頁點我開啟連結
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!
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
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