跳到主要內容

臺灣博碩士論文加值系統

(100.24.118.144) 您好!臺灣時間:2022/12/06 04:35
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

: 
twitterline
研究生:李翊豪
研究生(外文):LI,YI-HAO
論文名稱:整合熱像對稱性與可見光紋理特性之行人偵測
論文名稱(外文):Combination of Symmetry in Thermal and Texture in Visible Imagery for Pedestrian Detection
指導教授:黃世勳黃世勳引用關係
指導教授(外文):HUANG,SHIH-SHINH
口試委員:李素玲曾建誠許朝詠
口試委員(外文):LEE,SU-LINGTSENG,CHIEN-CHENGHSU,CHAO-YUNG
口試日期:2017-07-21
學位類別:碩士
校院名稱:國立高雄第一科技大學
系所名稱:電腦與通訊工程系碩士班
學門:工程學門
學類:電資工程學類
論文種類:學術論文
論文出版年:2017
畢業學年度:105
語文別:中文
論文頁數:57
中文關鍵詞:熱影像可見光整合行人偵測
外文關鍵詞:ThermalVisibleFusionPedestrian Detection
相關次數:
  • 被引用被引用:0
  • 點閱點閱:221
  • 評分評分:
  • 下載下載:22
  • 收藏至我的研究室書目清單書目收藏:0
行人偵測被廣泛應用於智慧交通系統或監控等多種不同領域,一般而言,基於可見光感測器於日間光線充足情況下,可提供有效之影像資訊,然於光線不足環境中則無法運作;熱影像感測器於夜間環境中,因無環境背景熱源干擾,可獲得顯著之行人資訊,然而,當行人與環境熱源相近時,通常為日間環境中,則效能會因而遞減,基於上述,本研究目的主要在提出一個結合熱影像與可見光之全天候行人偵測演算法。
本研究分別獨立對熱像與可見光感測器訓練出一組人形分類模型,並應用於後續之行人偵測,上述兩者之差異主要基於熱像與可見光影像特性之不同,進而分別採用不同之特徵,並結合支援向量機(Support Vector Machine)作為人形分類之依據。於熱像部分主要透過所提出之對稱權重HOG (SW-HOG:Symmetry-Weighted HOG)作為人形外觀特徵,其主要概念為整合HOG特徵於水平與垂直方式之對稱關係,以提升HOG特徵之有效鑑別度;可見光部分則以透過HOG及GoogLenet卷積式類神經網路(Convolution Neural Network)最後一層所獲得之表示式分別獲得影像中紋理與RGB色彩空間資訊,最後透過決策融合方式達到整合不同影像資訊之目的。
於實驗部分透過所建置之行人資料庫,其中包含四段靜態背景與兩段動態影像序列作為驗證演算法之依據,此外,並同時實作文獻中常用之行人偵測演算法,與所提出之演算法進行分析比較,於所獲得之FPPI曲線圖可發現所提出之演算法可提高文獻中行人偵測之準確度。
Pedestrian detection is widely used in intelligent transportation and surveillance systems. In general, the visible sensor is effective in daytime, but it fails in the poor lighting situation, such as nighttime. On the contrary, the thermal sensor can well sense the heat of the pedestrian body well in nighttime. But, it generally fails in case of that the temperature of background is close to pedestrian body, especially, in daytime. Accordingly, the objective of this study is to propose a pedestrian detection algorithm by fusing visible and thermal sensors so that it is suitable for both daytime and nighttime.
Since the visible and thermal sensors have different characteristics, the models for visible and thermal sensors are trained independently based on different features with Support Vector Machine (SVM). For thermal sensor, an effective feature called Symmetry-Weighted HOG (SW-HOG) feature is proposed to improve the discriminability of the traditional histogram of oriented gradients (HOG) feature. For visible sensor, the combination of HOG feature and the one extracting from GoogLenet
Convolution Neural Network (CNN) is used to model the texture properties of the pedestrian appearance. After that, the fusion mechanism is proposed to fuse the detection results from two sensors.
In experiment, six videos, four from stationary platform and two sequences from moving platform, are used to validate the proposed algorithm. Besides, several algorithms in the literature are implemented for comparison. Based on the FPPI analysis, the method proposed in this study outperforms the others

摘要 I
ABSTRACT II
誌謝 III
目錄 IV
圖目錄 V
表目錄 VII
第一章、緒論 1
1.1 研究動機 1
1.2 困難與挑戰 3
1.3 相關研究 4
1.4 系統概述(SYSTEM OVERVIEW) 6
第二章、人形分類器學習 8
2.1 HOG特徵與SVM簡介 8
2.2 基於SW-HOG之熱像人型分類器 11
2.3 基於紋理與色彩之可見光人型分類器 15
第三章、基於整合之行人偵測 19
3.1 HYPOTHESIS GENERATION 19
3.2 行人偵測 24
3.3 整合機制(FUSION MECHANISM) 25
第四章、實驗與分析 27
4.1 相機設置 27
4.2 訓練說明 28
4.3 測試驗證說明 30
4.4 效能分析 32
第五章、結論 45
參考文獻 46

[1]Piotr Dollár, Christian Wojek, Bernt Schiele, Pietro Perona: “Pedestrian Detection: An Evaluation of the State of the Art”, IEEE Trans. on Pattern Analysis and Machine Intelligence, 34(4), 743-761, 2012
[2]Rodrigo Benenson, Mohamed Omran, Jan Hendrik Hosang, Bernt Schiele: “Ten Years of Pedestrian Detection, What Have We Learned?”, ECCV Workshops, vol. 2, pp. 613-627, 2014
[3]Jan Thomanek, Marc Ritter, Holger Lietz, Gerd Wanielik, “Comparing Visual Data Fusion Techniques using FIR and Visible Light Sensors to Improve Pedestrian Detection,” Digital Image Computing Techniques and Applications, pp. 119-125, 2011
[4]Y. LeCun and Y. Bengio, “Convolutional networks for images, speech, and time series,” In M. Arbib, editor, The Handbook of Brain Theory and Neural Networks, pp. 255–258. MIT Press, Cambridge, Massachusetts, 1995
[5]Krizhevsky Alex, Ilya Sutskever, Geoffrey E. Hinton. “ImageNet classification with deep convolutional neural networks,” In NIPS, 2012
[6]Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., and Rabinovich, A. “Going deeper with convolutions,” CoRR, pp. 1409-4842, 2014
[7]Karen Simonyan, Andrew Zisserman, “Very Deep Convolutional Networks for Large-Scale Image Recognition,” CVPR, pp. 1409-1556, 2014
[8]Ankit Verma, Ramya Hebbalaguppe, Lovekesh Vig, Swagat Kumar, and Ehtesham Hassan, ” Pedestrian Detection via Mixture of CNN Experts and thresholded Aggregated Channel Features,” IEEE Intl. Conf. Computer Vision Workshop, pp. 555—563, 2015
[9]Dalal N., Triggs B., “Histograms of oriented gradients for human detection, ” In Proceedings of the IEEE conf. Computer Vision and Pattern Recognition, pp. 886–893, 2005
[10]C. Harris and M. Stephens. “A combined edge and corner detector,” Proceedings of the 4th Alvey Vision Conference, pp. 147–151, 1988
[11]D.S.Kim, M.Kim, B.S.Kim, K.H.Kim, K.H.Lee, “Histograms of local intensity differences for pedestrian classification in far-infrared images,” IEL Electronics Letters, vol.49, no.4, pp.258-260, 2013
[12]B.E.Boser, I.M.Guyon, V.N.Vapnik, "A training algorithm for optimal margin classifiers,” Proceedings of the fifth annual workshop on Computational learning theory, pp.144. 1992
[13]Piotr Dollár, Zhuowen Tu, Pietro Perona, Serge Belongie,” Integral Channel Features,” In: Proceedings of the British Machine Vision Conf. BMVC, pp.91.1-91.11, 2009
[14]Raluca Brehar, Cristian Vancea and Sergiu Nedevschi, “Pedestrian Detection in Infrared Images Using Aggregated Channel Features,” Intelligent Computer Communication and Processing(ICCP), pp.127-132, 2014
[15]Soonmin Hwang, Jaesik Park, Namil Kim, Yukyung Choi, In So Kweon “Multispectral Pedestrian Detection: Benchmark Dataset and Baseline,” In the IEEE conf. Computer Vision and Pattern Recognition(CVPR), pp. 1037–1045, 2015

QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top