跳到主要內容

簡易檢索 / 詳目顯示

回結果列表
研究生： 戴子傑
Tzjie Dai
論文名稱： 應用生成對抗網路於行車紀錄器影像之反射影去除演算法
The Application of the Generative Adversarial Network in the Reflection Removal Algorithm for Dashcam Images
指導教授： 蘇木春
Mu-Chun Su
口試委員:
學位類別： 碩士
Master
系所名稱： 資訊電機學院 - 資訊工程學系
Department of Computer Science & Information Engineering
論文出版年： 2020
畢業學年度： 108
語文別： 中文
論文頁數： 59
中文關鍵詞： 影像反射影去除 、生成對抗網路 、物件偵測
外文關鍵詞： Reflect Removal for Single image, Generative Adversarial Network, Object Detection
相關次數： 點閱：10  下載：0 
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報

    • 電腦視覺在近幾年高度發展，在物件偵測方面，YOLO、R-CNN 已經有不錯表現。如果輸入影像因拍攝時隔著玻璃，受到反射影干擾，這些物件偵測演算法會受到影響，導致定位失準或甚至遺漏。因此本論文希望提出一個基於生成對抗網路 (Generative Adversarial Network) 對單一影像進行反射影去除的演算法，來降低其影響，使得物件偵測效果能夠有所改善。
    本論文以行車紀錄器影像為主要應用情境，偵測包含汽車、機車、卡車、
    公車、腳踏車五種常見的交通工具。由於行車紀錄器常安裝於汽車擋風玻璃後，影像有機會受到玻璃上的反射影影響，進而使物件偵測效果變差，從而影響駕駛輔助系統。所以，本論文先訓練 Mask R-CNN，並用以偵測反射影及其位置，再訓練一個生成對抗網路，使其可以利用反射影後方的特徵進行還原，達到反射影去除的效果，藉此進一步改善物件偵測的準確度。

    Computer vision has been highly developed in recent years. In terms of object detection, YOLO and R-CNN have already performed well. However, if the input image is disturbed by reflections due to the separation of the glass during shooting, these object detection algorithms will be affected, resulting in misalignment or even omission. Therefore, this thesis tries to propose an algorithm based on Generative Adversarial Network (GAN) to remove the reflection of the single image to reduce its impact and improve the object detection effect.
    This thesis uses the dashcam image as the main application scenario to detect five common vehicles including cars, motorbikes, trucks, buses, and bicycles. Since the dashcam is usually installed behind the windshield of the car, the image may be affected by the reflection on it, which may make the accuracy of the object detection worse, thereby affecting the driving assistance system. Therefore, this thesis first trains a Mask R-CNN to detect the mask of the reflection, and then trains the GAN so that it can use the features behind the reflection to restore and achieve the effect of reflection removal. Further raises mAP of object detection.

    摘要 i ABSTRACT ii 致謝 iii 目錄 iv 圖目錄 vi 表目錄 viii 第一章、緒論 1 1-1 研究動機 1 1-2 研究目的 2 1-3 論文架構 3 第二章、相關研究 4 2-1 反射影去除 4 2-1-1 反射影類型簡介 5 2-1-2 反射影去除問題之輸入與輸出 6 2-1-3 現有方法 7 2-2 物件偵測 11 2-2-1 R-CNN 12 2-2-2 YOLO 14 2-2-3 Mask R-CNN 15 2-3 生成對抗網路 17 2-3-1 Conditional GAN 17 2-3-2 Pix2pix Conditional GAN 18 第三章、研究方法 20 3-1 演算法流程 20 3-2 反射影偵測 21 3-3 淡化反射影 22 3-4 利用物件偵測進行效果檢驗 22 第四章、實驗設計與結果 23 4-1 反射影偵測實驗 23 4-1-1 實驗設計 23 4-1-2 實驗結果 24 4-1-3 實驗分析 28 4-2 淡化反射影及GAN還原實驗 31 4-2-1 實驗設計 31 4-2-2 實驗結果 32 4-2-3 還原真實反射影 35 4-2-4 實驗分析 37 4-3 利用物件偵測進行效果檢驗實驗 39 4-3-1 實驗設計 39 4-3-2 實驗結果 39 4-3-3 實驗分析 40 第五章、結論與未來展望 42 5-1 結論 42 5-2 未來展望 43 參考文獻 44

    [1] "ImageNet," Stanford University; Princeton University, [Online]. Available: http://www.image-net.org/. [Accessed 6 July 2020].
    [2] A. Krizhevsky, I. Sutskever and G. E. Hinton, "ImageNet Classification with Deep Convolutional Neural Networks," in Neural Information Processing Systems, 2012.
    [3] D. Silver, A. Huang, C. J. Maddison, A. Guez, L. Sifre, G. v. d. Driessche, J. Schrittwieser, I. Antonoglou, V. Panneershelvam, M. Lanctot, S. Dieleman, D. Grewe, J. Nham, N. Kalchbrenner, I. Sutskever, T. Lillicrap, M. Leach, K. Kavukcuoglu, T. Graepel and D. Hassabis, "Mastering the game of Go with deep neural networks and tree search," Nature, vol. 529, pp. 484-489, 2016.
    [4] J. Redmon, S. Divvala, R. Girshick and A. Farhadi, "You Only Look Once: Unified, Real-Time Object Detection," in IEEE Conference on Computer Vision and Pattern Recognition, 2016.
    [5] R. Girshick, J. Donahue, T. Darrell and J. Malik, "Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation," in IEEE Conference on Computer Vision and Pattern Recognition, 2014.
    [6] K. He, G. Gkioxari, P. Dollár and R. Girshick, "Mask R-CNN," in IEEE International Conference on Computer Vision , 2017.
    [7] I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville and Y. Bengio, "Generative Adversarial Nets," in Proceedings of the Neural Information Processing Systems, 2014.
    [8] J. Redmon and A. Farhadi, "YOLOv3: An Incremental Improvement," in arXiv:1804.02767, 2018.
    [9] D. Lee, M.-H. Yang and S. Oh, "Generative Single Image Reflection Separation," arXiv:1801.04102, 2018.
    [10] R. Wan, B. Shi, L.-Y. Duan, A.-H. Tan and A. C. Kot, "Benchmarking Single-Image Reflection Removal Algorithms," in IEEE International Conference on Computer Vision (ICCV), 2017.
    [11] Q. Wen, Y. Tan, J. Qin, W. Liu, G. Han and S. He, "Single Image Reflection Removal Beyond Linearity," in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019.
    [12] M. Hu, "Reflection Removal Algorithms," 2016. [Online]. Available: http://stanford.edu/class/ee367/Winter2016/Hu_Report.pdf. [Accessed 20 06 2020].
    [13] Z. Chi, X. Wu, X. Shu and J. Gu, "Single Image Reflection Removal Using Deep Encoder-Decoder Network," arXiv:1802.00094, 2018.
    [14] R. Wan, B. Shi, L.-Y. Duan, A.-H. Tan, W. Gao and A. C. Kot, "Region-Aware Reflection Removal With Unified Content and Gradient Priors," IEEE Transactions on Image Processing, vol. 27, pp. 2927-2941, 2018.
    [15] Q. Fan, J. Yang, G. Hua, B. Chen and D. Wipf, "A Generic Deep Architecture for Single Image Reflection Removal and Image Smoothing," in arXiv:1708.03474, 2017.
    [16] X. Zhang, R. Ng and Q. Chen, "Single Image Reflection Separation with Perceptual Losses," in arXiv:1806.05376, 2018.
    [17] K. Wei, J. Yang, Y. Fu, D. Wipf and H. Huang, "Single Image Reflection Removal Exploiting Misaligned Training Data and Network Enhancements," in arXiv:1904.00637, 2019.
    [18] R. Fu, P. Kuang, Y. Zhou, H.-r. Yan and T.-Y. Zheng, "Area-Aware Reflection Detection and Removal for Single Image," in 16th International Computer Conference on Wavelet Active Media Technology and Information Processing, 2019.
    [19] F.-F. Li, J. Johnson and S. Yeung, "Stanford University CS231n: Convolutional Neural Networks for Visual Recognition," 14 May 2019. [Online]. Available: http://cs231n.stanford.edu/slides/2019/cs231n_2019_lecture12.pdf. [Accessed 20 June 2020].
    [20] R. Girshick, "Fast R-CNN," in IEEE International Conference on Computer Vision, 2015.
    [21] S. Ren, K. He, R. Girshick and J. Sun, "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 6, pp. 1137-1149, 2016.
    [22] J. Redmon and A. Farhadi, "YOLO9000: Better, Faster, Stronger," in arXiv:1612.08242, 2016.
    [23] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C. Y. Fu and A. C. Berg, "SSD: Single Shot MultiBox Detector," in Computer Vision and Pattern Recognition, 2016.
    [24] C. Cortes and V. Vapnik, "Support-vector network," Machine learning, vol. 20, no. 3, pp. 273-297, 1995.
    [25] K. He, X. Zhang, S. Ren and J. Sun, "Deep Residual Learning for Image Recognition," in arVix:1512.03385, 2015.
    [26] T. Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan and S. Belongie, "Feature Pyramid Networks for Object Detection," in Computer Vision and Pattern Recognition, 2017.
    [27] M. Mirza and S. Osindero, "Conditional Generative Adversarial Nets," in arXiv, 2014.
    [28] P. Isola, J.-Y. Zhu, T. Zhou and A. A. Efros, "Image-to-Image Translation with Conditional Adversarial Networks," in Computer Vision and Pattern Recognition, 2017.
    [29] "OpenCV," [Online]. Available: https://opencv.org/. [Accessed 20 June 2020].
    [30] "Nexar Challenge 2," Nexar, [Online]. Available: https://www.getnexar.com/challenge-2/. [Accessed 24 May 2019].
    [31] "Labelme: Image Polygonal Annotation with Python," [Online]. Available: https://github.com/wkentaro/labelme. [Accessed 20 June 2020].

    QR CODE
    :::