現今，自駕車定位技術的發展依舊是一個很大的挑戰，被廣泛使用的全球導航衛 星系統(GNSS)雖可達到次公尺等級的定位準確率，但在特定場景下，衛星訊號有可能 被遮蔽或反射，這會導致定位結果不穩定，所以開發不同感測器的定位技術是勢在必行 的。此外，由於自駕車將面臨市場化及商業化的需求，所以感測器的成本及便攜性將成 為發展定位策略的考量重點之一，例如，因為製造工業的進步，慣性感測單元的體積縮 小、成本降低，用於定位方面的研究隨之蓬勃發展，但其積分誤差會隨時間逐漸累積， 故需要其他感測器輔助，相機及光達皆可提供豐富的環境資訊，所以也被廣泛用於自駕 車輛的技術發展，但由於光達造價昂貴，故而此篇論文將選用相機作為周遭環境的感測 器，與慣性感測器融合估計車子的位置及姿態。
感測器融合的方法主要可分為兩種，其一為利用最佳化的方法進行迭代，求出最 小化定位誤差，此方法耗費較多時間及運算資源，但也能獲得較精確的定位解;另一為 基於濾波器的方法，利用機率分布建構運動模型，對狀態進行估計，估計出來的狀態雖 不比最佳化法精確，但時間及運算成本較低，故若考慮車輛動態及影像處理的時間消耗， 此篇論文選用擴展卡爾曼濾波器來完成位置估計。此外，隨著對定位準確度要求的提高，高精度地圖隨之產生，以提供公分等級精確度的道路訊息為其優勢，也將應用於此份研究之中。
本論文提出將高精地圖加入慣性感測單元及相機的定位演算法架構，藉由將高精 地圖中的柱狀特徵投影到照片上來提取感興趣區域(ROI)，而後僅對此區域進行影像處 理，可降低整體計算量，提升效率。此外，將高精度地圖中柱狀特徵的三維座標加入測 量模型中，提高測量狀態的可靠度。最後，將此架構透過 Autoware 開源軟體實現台南 高鐵站高架鐵路周圍的定位計算，並將結果進行討論與分析。

Nowadays, Localization is still a challenge for highly automated vehicles (AVs). The well-known Global Navigation Satellite System (GNSS) can provide submeter level accuracy for AVs. However, the signal outage of GNSS causes unstable positioning in some scenarios. Hence, developing localization methods by combining other sensors is needed.
Besides, in recent years, automated vehicles are trying to reach the market, so the discontinuous performance of localization and high cost of sensors are not allowed. The inertial sensors, e.g. Inertial Measurement Unit (IMU), have been applied, but the integration error of this kind of sensor will grow drastically over time. Moreover, Lidar is too expensive to be suitable for commercial vehicles, and its volume is larger than the camera, a candidate replacement. As a result, we choose a camera and IMU to estimate the attitude and position of the vehicle.
Concerning the sensor fusion methods, they can be roughly categorized into two types, optimization-based and filter-based. Optimization-based methods often use an iterative calculation to solve the problem. They sometimes take a bunch of time and use too many computing resources, but if they finish, they can get a more accurate solution. On the other hand, filter-based methods get a less accurate solution, but the time consumption would not be a problem. The extended Kalman filter is used in this work since the feature detection process would take some time. Nevertheless, because the localization aims to achieve lane-level accuracy these years, the High Definition (HD) map has been frequently applied to AVs.
By reprojecting the pole features in the HD map to the camera image plane, the ROI can be generated. With this process, feature detection becomes more efficient. Additionally, cooperated with the 3D absolute position of the features in the HD map, the measurement model can get a more precise error estimation. We implement it in Autoware and discuss the results.

[1] Y. Alkendi, L. Seneviratne, and Y. Zweiri, "State of the art in vision-based localization techniques for autonomous navigation systems," IEEE Access, vol. 9, pp. 76847-76874, 2021, doi: 10.1109/ACCESS.2021.3082778.
[2] N. Piasco, D. Sidibé, C. Demonceaux, and V. Gouet-Brunet, "A survey on visual-based localization: On the benefit of heterogeneous data," Pattern Recognition, vol. 74, pp. 90- 109, 2018.
[3] C. Cimarelli, D. Cazzato, M. A. Olivares-Mendez, and H. Voos, "Faster visual-based localization with mobile-posenet," in International Conference on Computer Analysis of Images and Patterns, 2019: Springer, pp. 219-230.
[4] R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, "Grad- cam: Visual explanations from deep networks via gradient-based localization," in Proceedings of the IEEE International Conference on Computer Vision, 2017, pp. 618- 626.
[5] K. Khoshelham and M. Ramezani, "Vehicle positioning in the absence of GNSS signals: Potential of visual-inertial odometry," in 2017 Joint Urban Remote Sensing Event (JURSE), 2017: IEEE, pp. 1-4.
[6] S. Leutenegger, P. Furgale, V. Rabaud, M. Chli, K. Konolige, and R. Siegwart, "Keyframe-based visual-inertial slam using nonlinear optimization," in Proceedings of Robotis Science and Systems (RSS) 2013, 2013.
[7] T. Qin, P. Li, and S. Shen, "VINS-Mono: A robust and versatile monocular visual- inertial state estimator," IEEE Transactions on Robotics, vol. 34, no. 4, pp. 1004-1020, 2018, doi: 10.1109/TRO.2018.2853729.
[8] R. E. Kalman, "A new approach to linear filtering and prediction problems," Journal of Basic Engineering, vol. 82, no. 1, pp. 35-45, 1960, doi: 10.1115/1.3662552.
[9] E. Stenborg, C. Toft, and L. Hammarstrand, "Long-term visual localization using semantically segmented images," in 2018 IEEE International Conference on Robotics and Automation (ICRA), 2018: IEEE, pp. 6484-6490.
[10] M. Ramezani and K. Khoshelham, "Vehicle positioning in GNSS-deprived urban areas by stereo visual-inertial odometry," IEEE Transactions on Intelligent Vehicles, vol. 3, no. 2, pp. 208-217, 2018.
[11] M. Schreiber, H. Königshof, A.-M. Hellmund, and C. Stiller, "Vehicle localization with tightly coupled GNSS and visual odometry," in 2016 IEEE Intelligent Vehicles
Symposium (IV), 2016: IEEE, pp. 858-863.
[12] M. Brossard, S. Bonnabel, and A. Barrau, "Unscented Kalman filtering on Lie groups for fusion of IMU and monocular vision," in Proc. Int. Conf. Robot. Automat.(ICRA), 2017, pp. 1-9.
[13] A. I. Mourikis and S. I. Roumeliotis, "A multi-state constraint Kalman filter for vision-aided inertial navigation," in Proceedings 2007 IEEE International Conference on Robotics and Automation, 10-14 April 2007 2007, pp. 3565-3572, doi: 10.1109/ROBOT.2007.364024.
[14] "Creating maps for Autoware - Autoware Documentation." autowarefoundation.github. io. https://autowarefoundation.github.io/autoware-documentation/main/how-to-guides/ creating-maps-for-autoware/#vector-maps (accessed Aug. 8, 2022).
[15] W. Liu, D. Song, Z. Wang, and K. Fang, "Comparative Analysis between Error-State and Full-State Error Estimation for KF-Based IMU/GNSS Integration against IMU Faults," Sensors, vol. 19, no. 22, p. 4912, 2019.
[16] N. Trawny and S. I. Roumeliotis, "Indirect Kalman filter for 3D attitude estimation," University of Minnesota, Dept. of Comp. Sci. & Eng., Tech. Rep, vol. 2, p. 2005, 2005.
[17] "OpenCV: Camera Calibration." opencv.org. https://docs.opencv.org/4.x/dc/dbb/tutoria l_py_calibration.html (accessed Jul. 23, 2022).
[18] D. H. Ballard, "Generalizing the Hough transform to detect arbitrary shapes," Pattern Recognition, vol. 13, no. 2, pp. 111-122, 1981.
[19] C. Galamhos, J. Matas, and J. Kittler, "Progressive probabilistic Hough transform for line detection," in Proceedings. 1999 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (Cat. No PR00149), 1999, vol. 1: IEEE, pp. 554-560.
[20] "OEM7 SPAN Overview." novatel.com. https://docs.novatel.com/OEM7/Content/SPA N_Install/Introduction.htm?tocpath=Installation%7COEM7%20SPAN%20Overview% 7C_____0 (accessed Jul. 24, 2022).