臺灣博碩士論文加值系統

近年全球自駕車技術如雨後春筍般的冒出，現階段的技術大多都已進入了道路測試階段，像是自駕巴士、園區自駕接駁車等，因此自駕車的定位對於行車安全是一個相當重要的能力之一，而 AMCL（Adaptive Monte Carlo Localization）演算法是常用的定位方法，但此演算法若在初始位置發生定位錯誤，會引發交通安全的問題。因此本研究利用CNN（Convolutional Neural Network）模型來辨識自駕車前方的特殊景色，為AMCL 提供一個接近實際位置的參考座標，進而讓 AMCL 的初始粒子散佈在其座標周圍，故得以快速收斂在正確的位置上。本研究先提出路段景像定位法，雖然可以解決此迷航問題，但是由於需事先經驗區分路段，導致人工成本太高。為了改善此問題，我們利用ORB（Oriented FAST and Rotated BRIEF）萃取路段上的特徵點，並利用形態學將群聚的特徵點結合為特徵物件，隨後利用二維向量來描述此特徵物件的形狀及主軸角，最後會根據此向量的匹配對是否超過預設比率來決定路段的長度，進而解決因人工所擇選的固定長度路段，導致人工成本過高之問題。
特徵點的萃取對於影像匹配是一個重要的角色，若特徵點萃取過少，會造成匹配點對數量過少，導致匹配失敗或是匹配的準確度降低。為了解決此問題，本研究提出飽和點極性演算法加快找出飽和色彩，然後透過調整影像色彩的飽和度，來增加影像色彩對比，進一步讓ORB 演算法在特徵點檢測時提升萃取到的特徵點數量。
綜上，本論文提出路段景像、路段特徵物件及快速強化色彩飽和對比等技術，可增強影像特徵並解決自駕車的迷航問題，進而強化了行車安全。

In this study, a CNN(Convolutional Neural Network) model is used to identify the special scenery in front of the self­-driving vehicle to provide a reference coordinates close to the actual position of the AMCL(Adaptive Monte Carlo Localization), which enables the initial particles of the AMCL to be scattered around its coordinates, so that it can quickly converge on the cor­rect position. we first proposed the road segment view localization method, which can solve the lost navigation problem, but the human cost is too high due to the prior experience required to distinguish the road segments. To improve this problem, we use ORB(Oriented FAST and Rotated BRIEF) to extract the feature points on the road segments and use morphology to com­ bine the clustered feature points into feature objects. Then, a two-­dimensional vector is used to describe the shape and axis angle of this feature objects, and finally, the length of the section is determined by whether the matching pair of this vector exceeds the default ratio. The problem of high human cost due to the fixed length of the road chosen can be solved.
To solve the problem of too few feature points extracted, resulting in too few matching pairs, this paper proposes a saturation point polarity algorithm to speed up finding saturated colors and increase the image color contrast by adjusting the image color saturation, so that the ORB algorithm can increase the number of feature points extracted during feature point detection.
In summary, this paper proposes techniques such as road view, road feature objects and rapid enhancement of color saturation contrast to enhance image features and solve the naviga­tion problem of self­-driving vehicles, which further enhances driving safety.

摘要 i
ABSTRACT ii
致謝 iii
目錄 iv
表目錄 vi
圖目錄 vii
第一章 緒論 1
1.1 前言 1
1.2 研究動機與目的 2
1.3 論文架構 3
第二章 相關研究 4
2.1 自駕車定位 4
2.2 卷積神經網路 6
2.2.1 卷積層 7
2.2.2 池化層 8
2.2.3 全連接層 8
2.3 特徵點偵測 10
第三章 實驗環境介紹 12
3.1 Ubuntu 作業系統 12
3.2 Robot Operating System 13
3.3 TensorFlow 14
3.4 ESC8000 G3 16
3.5 自駕車設備 18
第四章 實驗設計 23
4.1 運用影像相似度辨別自駕車所在地的位置資訊 23
4.1.1 路段區分 23
4.1.2 數據集（Dataset） 24
4.1.3 實驗模型 27
4.1.4 實驗結果 28
4.2 利用影像中的特徵物件產生自適應基準影像 32
4.2.1 ROI 定義 33
4.2.2 特徵物件 35
4.2.3 特徵物件向量 36
4.2.4 匹配 37
4.2.5 產生新的基準畫面 38
4.2.6 實驗結果 39
4.3 強化色彩對比增進特徵點強健性 41
4.3.1 色彩空間 41
4.3.2 飽和點極性演算法 42
4.3.3 微飽和影像 46
4.3.4 實驗結果 47
第五章 結論與未來工作 48
5.1 結論 48
5.2 未來工作 49
參考文獻 50

[1] D. Fox, “Kld­sampling: Adaptive particle filters,” in Advances in Neural Information Processing Systems, T. Dietterich, S. Becker, and Z. Ghahramani, Eds., vol. 14, MIT Press, 2001. [Online]. Available: https://proceedings.neurips.cc/paper/2001/file/c5b2cebf15b205503560c4e8e6d1ea78-Paper.pdf.
[2] F. Dellaert, D. Fox, W. Burgard, and S. Thrun, “Monte carlo localization for mobile robots,” in Proceedings 1999 IEEE International Conference on Robotics and Automation (Cat. No.99CH36288C), vol. 2, 1999, 1322–1328 vol.2. DOI: 10 . 1109 / ROBOT .1999.772544.
[3] Particle filter. [Online]. Available: https://en.wikipedia.org/wiki/Particle_filter.
[4] Convolutional neural network. [Online]. Available: https://en.wikipedia.org/wiki/Convolutional_neural_network.
[5] F. K. Noble, “Comparison of opencv’s feature detectors and feature matchers,” in 2016 23rd International Conference on Mechatronics and Machine Vision in Practice (M2VIP),2016, pp. 1–6. DOI: 10.1109/M2VIP.2016.7827292.
[6] E. Rublee, V. Rabaud, K. Konolige, and G. Bradski, “Orb: An efficient alternative to sift or surf,” in 2011 International Conference on Computer Vision, 2011, pp. 2564–2571.DOI: 10.1109/ICCV.2011.6126544
[7] D. G. Lowe, “Distinctive image features from scale­invariant keypoints,” International journal of computer vision, vol. 60, no. 2, pp. 91–110, 2004.
[8] H. Bay, “Surf: Speeded up robust features,” pp. 404–417, 2006.
[9] K.­L. Chung, W.­J. Yang, and W.­M. Yan, “Efficient edge­preserving algorithm for color contrast enhancement with application to color image segmentation,” Journal of Visual Communication and Image Representation, vol. 19, no. 5, pp. 299–310, 2008, ISSN: 1047­3203. DOI: https://doi.org/10.1016/j.jvcir.2008.02.002. [Online]. Available:https://www.sciencedirect.com/science/article/pii/S1047320308000229.
[10] Ubuntu 基於 debian 的開源操作系統, [Accessed 05 2022]. [Online]. Available: https://zh.m.wikipedia.org/zh-tw/Ubuntu.
[11] The ubuntu lifecycle and release cadence, 2022. [Online]. Available: https://ubuntu.com/about/release-cycle#ubuntu.
[12] Ros introduction, [Accessed 08 2018]. [Online]. Available: http://wiki.ros.org/ROS/Introduction.
[13] Tensorflow. [Online]. Available: https://zh.wikipedia.org/zh-tw/TensorFlow.
[14] 端對端的開放原始碼機器學習平台. [Online]. Available: https://www.tensorflow.org/?hl=zh-tw.
[15] J. Hale, Deep learning framework power scores 2018. [Online]. Available: https://towardsdatascience.com/deep- learning- framework- power- scores- 2018-23607ddf297a (visited on 09/20/2018).
[16] 林大貴, TensorFlow + Keras 深度學習人工智慧實務應用. 博碩文化股份有限公司,2017, ISBN: 9789864342167.
[17] AutonomouStuff, Velodyne puck. [Online]. Available: https : / / autonomoustuff.com/products/velodyne-puck-vlp-16.
[18] I. Corporation, Intel® realsense™ depth camera d435i. [Online]. Available: https://store.intelrealsense.com/buy- intel- realsense- depth- camera- d435i.html.
[19] 鍾國亮, 影像處理與電腦視覺, ser. Image Processing and computer Vision. 台灣東華書局股份有限公司, 2015, ISBN: 9789574838271.
[20] S.­C. Pei, Y.­C. Zeng, and C.­H. Chang, “Virtual restoration of ancient chinese paintings using color contrast enhancement and lacuna texture synthesis,” IEEE Transactions on Image Processing, vol. 13, no. 3, pp. 416–429, 2004. DOI: 10.1109/TIP.2003.821347.