由於近年來影像辨識能力的增強以及深度學習的技術的發展，車速辨識系統逐漸以一般攝影機進行偵測，我們只使用一台攝影機進行車速的辨識以降低硬體成本，並且在畫面中立即的計算車速而非計算一段距離之平均車速，以及透過影像中已知的資訊進行校正，使得裝置的安裝更為便利。
我們使用機器學習方式偵測車輛物件，不致於因車輛行駛的方向而受到限制，另外利用道路交通標線規則可以取得所有標線的大小，將其與影像中道路標線進行透視轉換得到校正矩陣，因此不需事先量測道路的大小或是得到影像與實際場景的比例尺即可完成校正，爾後完成車速辨識的功能。本篇論文以車速20km/hr、30km/hr以及40km/hr進行車速偵測，其車速平均誤差為2.19 km/hr，平均相對誤差為7.56%。
在未來的研究，我們除了必須研究如何提高車速辨識的準確性之外，還可以將系統輕量化，使得系統可建置於嵌入式裝置，讓系統建置成本更為低廉。

Based on recent developments in speed estimation and deep learning technology, we prefer to use only one camera for speed estimation to reduce hardware costs. We calculate the speed immediately on the screen instead of calculating the average speed over a distance and calibrate the video by known information immediately, making it easier to install the device.
We use a machine learning method to detect vehicle objects, so as not to be limited by the direction of the vehicle. In addition, the road traffic marking rules can be used to obtain the size of all the road markings, which can be converted to a correction matrix by road marking-based perspective transformation. Therefore, there is no need to measure the size of the road or get the scale of the image beforehand to calibrate and then complete the speed estimation. In this paper, speed detection is performed at 20km/hr, 30km/hr and 40km/hr. An average error is 2.19 km/hr, average relative error is 7.56%.
In the future, we plan to improve the recognition accuracy and make the system lighter and cheaper for building it in an embedded system.

[1] "警政統計查詢網." https://ba.npa.gov.tw/npa/stmain.jsp?sys=220&ym=10500&ymt=10800&kind=21&type=1&funid=q06020201&cycle=4&outmode=0&compmode=0&ohtml=q250x&outkind=1&fld0=1&cod00=1&cod10=1&rdm=a5exipel (accessed 3 July, 2020).
[2] D. Felguera-Martin, J. T. Gonzalez-Partida, P. Almorox-Gonzalez, and M. Burgos-García, "Vehicular Traffic Surveillance and Road Lane Detection Using Radar Interferometry," IEEE Transactions on Vehicular Technology, vol. 61, no. 3, pp. 959-970, 2012, doi: 10.1109/tvt.2012.2186323.
[3] A. E. C. Seas, MD, US), Field, Christopher (Baltimore, MD, US), Sirota, Marcos J. (Takoma Park, MD, US), "Compact single lens laser system for object/vehicle presence and speed determination," United States Patent Appl. 7323987, 2008. [Online]. Available: https://www.freepatentsonline.com/7323987.html
[4] S. Javadi, M. Dahl, and M. I. Pettersson, "Vehicle speed measurement model for video-based systems," Computers & Electrical Engineering, vol. 76, pp. 238-248, 2019, doi: 10.1016/j.compeleceng.2019.04.001.
[5] L. Yu, D. Zhang, X. Chen, and A. Hauptmann, "Traffic Danger Recognition With Surveillance Cameras Without Training Data," p. arXiv:1811.11969. [Online]. Available: https://ui.adsabs.harvard.edu/abs/2018arXiv181111969Y
[6] S. Hua, M. Kapoor, and D. C. Anastasiu, "Vehicle Tracking and Speed Estimation from Traffic Videos," presented at the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2018.
[7] L. Yang, M. Li, X. Song, Z. Xiong, C. Hou, and B. Qu, "Vehicle Speed Measurement Based on Binocular Stereovision System," IEEE Access, vol. 7, pp. 106628-106641, 2019, doi: 10.1109/access.2019.2932120.
[8] W. Czajewski and M. Iwanowski, "Vision-Based Vehicle Speed Measurement Method," in Computer Vision and Graphics, (Lecture Notes in Computer Science, 2010, ch. Chapter 35, pp. 308-315.
[9] D. C. Luvizon, B. T. Nassu, and R. Minetto, "A Video-Based System for Vehicle Speed Measurement in Urban Roadways," IEEE Transactions on Intelligent Transportation Systems, pp. 1-12, 2016, doi: 10.1109/tits.2016.2606369.
[10] V. Kocur and M. Ftáčnik, Detection of 3D Bounding Boxes of Vehicles Using Perspective Transformation for Accurate Speed Measurement. 2020.
[11] Y. Lecun, "Gradient-based learning applied to document recognition," Proceedings of the IEEE, vol. 86, no. 11, pp. 2278-2324, 1998 1998, doi: 10.1109/5.726791.
[12] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, "You Only Look Once: Unified, Real-Time Object Detection," presented at the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016.
[13] W. Liu et al., "SSD: Single Shot MultiBox Detector," Cham, 2016: Springer International Publishing, in Computer Vision – ECCV 2016, pp. 21-37.
[14] K. Simonyan and A. Zisserman, "Very deep convolutional networks for large-scale image recognition," arXiv preprint arXiv:1409.1556, 2014.
[15] M. Jaderberg, K. Simonyan, A. Zisserman, and K. Kavukcuoglu, "Spatial Transformer Networks," ArXiv, vol. abs/1506.02025, 2015.
[16] J. Redmon and A. Farhadi, "YOLO9000: Better, Faster, Stronger," presented at the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017.
[17] J. Mezirow, "Perspective Transformation," Adult Education, vol. 28, no. 2, pp. 100-110, 1978, doi: 10.1177/074171367802800202.
[18] S. M. Silva and C. R. Jung, "License Plate Detection and Recognition in Unconstrained Scenarios," Cham, 2018: Springer International Publishing, in Computer Vision – ECCV 2018, pp. 593-609.
[19] S. Montazzolli and C. Jung, "Real-Time Brazilian License Plate Detection and Recognition Using Deep Convolutional Neural Networks," presented at the 2017 30th SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI), 2017.