現今 ADAS 發展趨勢大部分多以使用多種感測器(LiDAR、Radar、GPS)收集環境
資訊，再依靠機器學習等高計算成本之演算法進行處理，但是 LiDAR 等感測器成本
與單目攝像機相比之下高昂許多，且高計算成本之演算法容易造成電腦之計算負擔，
造成可能無法作出及時之反應。本文提出以單相機立體視覺做為核心，構建出以自身
為原點之 3D 網格用以定義環境中立體資訊，結合分水嶺演算法(Watershed
Segmentation)以切割出開闊空間；並建立起一穩定流場系統以推算出穩定的移動視覺
擴張焦點(FoE)，再藉由觀察移動視覺擴張焦點座標變化推測出自身姿態之規則性
(Rule Based)系統。本文將使用德國卡爾斯魯厄理工學院和豐田美國技術研究院
(Karlsruhe Institute of Technology and Toyota Technological Institute, KITTI)聯合創辦的
計算機視覺演算法評測資料庫中鄉下道路、市區道路、州際公路等環境做驗證，證明
研究內容可行性。實驗結果顯示本文所提出透過結合單眼立體視覺與流場演算法之
Rule Based 演算法在白天駕駛環境能有效的估計出當前車輛姿態，並能夠在部分非複
雜駕駛環境中展現出不俗於部分使用機器學習/深度學習演算法之自身車輛姿態估
計，並且在系統穩定運行之狀況下，偏行角度(Yaw)最低誤差可到 0.0005 Degree/meter。

Most of the current ADAS development trend is to use multiple sensors (LiDAR, Radar,
GPS) to collect environmental information, and then rely on algorithms with high
computational cost such as machine learning for processing. However, the cost of sensors
such as LiDAR are much higher than monocular cameras, and high computational costs
algorithms might cause computational burdens on the computer. As a result, it may not be
possible to respond in a timely manner. This study proposes a rule-based system that takes
stereopsis approximation with single camera as the core, constructs a 3D grid to define the
stereo information in the environment, and cuts out the open space with the watershed
segmentation; creates a stable optical flow system to calculate the Focus of Expansion(FoE),
and estimates vehicle's posture by observing the coordinate changes of the FoE. This study
extracts suburban, urban and interstate highway, three data sets from the Karlsruhe Institute
of Technology in Germany and Toyota Technological Institute (KITTI) for verification, and
proves the feasibility of this system. The experimental results show that the rule-based
algorithm effectively estimates vehicle pose in daytime with monocular stereo vision and
optical flow, and show that our rule-based algorithm’s performance in non-complex driving
environment is better than some algorithms using machine learning/deep learning. Under the
condition in-non-complex driving environment, the minimum error of yaw reach 0.0005
degree/meter while proves the feasibility of proposed rule-based stereopsis.

摘要 ii
目錄 iv
圖目錄 vi
表目錄 viii
縮寫列表 ix
符號說明 x
壹、 緒論 1
一、 簡介 1
二、 貢獻 2
三、 引文 4
貳、研究與方法 6
一、 視線消失點(Vanishing Point)偵測 6
(一). Canny 邊緣偵測 7
(二). 霍夫轉換(Hough Transform) 7
(三). 視線消失點偵測(Vanish Point Detector) 7
二、 單眼立體視覺(Stereopsis with Single Camera) 8
(一). 格線建立 9
(二). LBP AUTO Marker 9
(三). Watershed 10
(四). 3D_Bounding_Box 10
三、 環景系統 11
(一). Bilinear interpolation 12
(二). Array Panorama 12
四、 流場建立 13
(二). 緊密流場－Farnebäck Optical Flow Algorithm 14
(三). 稀疏流場 － Harris Orb Matching Flow 15
(四). 緊密流場過濾器 15
(五). 稀疏流場過濾器 16
五、 移動視覺擴張焦點(Focus of Expansion, FoE)偵測 17
(一). FoE Detector 18
(三). FoE Filter 19
六、 同步定位與地圖構建系統(Simultaneous Localization and Mapping , SLAM) 20
(一). 透視投影(Perspective Projection) 21
(二). 姿態偵測 21
參、實驗成果展示 23
一、 3D Bounding Box 23
二、 車體姿態偵測 24
(一). 深度偵測 25
(二). 姿態偵測 26
肆、結論 32
一、 研究總結 32
二、 未來展望 32
伍、 參考文獻 34
陸、 附錄 36

[1]楊豐瑞, "單鏡頭前方視覺於先進駕駛輔助系統之應用," 碩士論文, 國立高雄科技大學電子工程系(第一校區), 2020.
[2]R. Jain, "Direct Computation of the Focus of Expansion," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. PAMI-5, no. 1, pp. 58-64, Jan. 1983.
[3]D. H. Ballard, "Generalizing the Hough transform to detect arbitrary shapes," Pattern Recognition, vol. 13, issue 2, pp. 111-122, Elsevier, 1981.
[4]G. Welch and G. Bishop, "An introduction to the Kalman filter," 1995. https://perso.crans.org/club-krobot/doc/kalman.pdf
[5]G. Farnebäck, "Two-frame motion estimation based on polynomial expansion," Scandinavian conference on Image analysis, pp. 363-370, Springer, Berlin, Heidelberg, 2003.
[6]Hieu Tat Nguyen, M. Worring and R. van den Boomgaard, "Watersnakes: energy-driven watershed segmentation," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 25, no. 3, pp. 330-342, March 2003.
[7]K. G. Derpanis, "The harris corner detector," York University 2, 2004. https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.482.1724&rep=rep1&type=pdf
[8]P. Bao, L. Zhang, and X. Wu, "Canny edge detection enhancement by scale multiplication,"IEEE transactions on pattern analysis and machine intelligence, vol. 27,no. 9, pp. 1485-1490, Sept 2005.
[9]A. J. Davison, I. D. Reid, N. D. Molton and O. Stasse, "MonoSLAM: Real-Time Single Camera SLAM," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 29, no. 6, pp. 1052-1067, June 2007.
[10]H. Badino, U. Franke, and D. Pfeiffer, "The stixel world-a compact medium level representation of the 3d-world," in Joint Pattern Recognition Symposium, pp. 51-60, Springer, 2009.
[11]H. Kong, J.-Y. Audibert, and J. Ponce, "Vanishing point detection for road detection," 2009 IEEE Conference on Computer Vision and Pattern Recognition, pp. 96-103, 2009.
[12]M. Calonder, V. Lepetit, M. Ozuysal, T. Trzcinski, C. Strecha, and P. Fua, "BRIEF: Computing a local binary descriptor very fast," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 34, no. 7, pp. 1281-1298, July 2012 .
[13]S. Li, C. Xu, and M. Xie, "A robust O (n) solution to the perspective-n-point problem," in IEEE transactions on pattern analysis and machine intelligence, vol. 34, no. 7, pp. 1444-1450, July 2012.
[14]A. Geiger, P. Lenz, C. Stiller, and R. Urtasun, "Vision meets robotics: The kitti dataset," in The International Journal of Robotics Research, vol. 32, no. 11, pp. 1231-1237, 2013.
[15]R. Mur-Artal, J. M. M. Montiel and J. D. Tardós, "ORB-SLAM: A Versatile and Accurate Monocular SLAM System," in IEEE Transactions on Robotics, vol. 31, no. 5, pp. 1147-1163, Oct. 2015
[16]J. D. Adarve and R. Mahony, "A Filter Formulation for Computing Real Time Optical Flow," in IEEE Robotics and Automation Letters, vol. 1, no. 2, pp. 1192-1199, July 2016.
[17]L. Sevilla-Lara, D. Sun, V. Jampani and M. J. Black, "Optical Flow with Semantic Segmentation and Localized Layers," 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pp. 3889-3898, 2016.
[18]E. Karami, S. Prasad, and M. Shehata, "Image matching using SIFT, SURF, BRIEF and ORB: performance comparison for distorted images," arXiv preprint arXiv:1710.02726, 2017.
[19]N. Fanani, A. Stürck, M. Barnada and R. Mester, "Multimodal scale estimation for monocular visual odometry," 2017 IEEE Intelligent Vehicles Symposium (IV) , pp. 1714-1721, 2017.
[20]C. Godard, O. M. Aodha, M. Firman and G. Brostow, "Digging Into Self-Supervised Monocular Depth Estimation," 2019 IEEE/CVF International Conference on Computer Vision (ICCV), pp. 3827-3837, 2019.
[21]Jiawang Bian, Zhichao Li, Naiyan Wang, Huangying Zhan, Chunhua Shen, Ming-Ming Cheng, Ian Reid, "Unsupervised Scale-consistent Depth and Ego-motion Learning from Monocular Video," Advances in Neural Information Processing Systems, vol. 32, 2019.
[22]Yuliang Zou , Pan Ji , Quoc-Huy Tran , Jia-Bin Huang , and Manmohan Chandraker, " Learning monocular visual odometry via self-supervised long-term modeling," in European Conference on Computer Vision, pp. 1231-1237, Springer, 2020.
[23] Z. Boukhers, K. Shirahama and M. Grzegorzek, "Example-Based 3D Trajectory Extraction of Objects From 2D Videos," in IEEE Transactions on Circuits and Systems for Video Technology, vol. 28, no. 9, pp. 2246-2260, Sept. 2018.
[24] Riley, Kenneth Franklin, Michael Paul Hobson, and Stephen John Bence, "Mathematical Methods for Physics and Engineering", 1999.