臺灣博碩士論文加值系統

在許多先進國家使用的智慧運輸系統(Intelligent Transportation System, ITS)，針對如何降低交通壅塞的問題，車速的估算是一個重要關鍵之一，因此，得知即時與準確的車速對於改善交通有很大的助益。本論文使用深度學習演算法 YOLOv5與感興趣區域(Region of Interest, ROI)等電腦視覺技術進行高速公路之車輛物件偵測、追蹤與估算車速，將高速公路各車道進行車輛計數與車速估算，以保證各車道的車輛能夠準確地被偵測到，而在物件偵測時，當車輛中心點進入ROI範圍的起始線時，開始進行追蹤，直到車輛離開ROI範圍的終點線，才停止追蹤，利用記錄進出ROI範圍的位置與時間進行車速估算；並以BrnoCompSpeed數據集中提供的影片，從中選取車輛速度快、速度中與速度慢三種不同場景進行驗證，同時以平均車速、標準差、變異數與準確率，成功地驗證所提方法之準確度與有效性。另外，本論文也與其他方法進行比較，以平均誤差、中位數誤差、平均精確率、平均召回率及處理速度，來說明所提方法所在速度與準確度上的優劣勢。最後本論文使用交通部高速公路局所提供的高速公路即時影像，證明所提方法的可用性與有效性。

Intelligent Transportation Systems (ITS) are used to address traffic congestion issues in many advanced countries. One critical aspect of mitigating congestion is the estimation of vehicle speed. Accurate and real-time speed data significantly contribute to traffic improvement. This thesis employs deep learning algorithms, specifically YOLOv5, along with computer vision techniques such as Region of Interest (ROI) to detect, track, and estimate vehicle speeds on highways. By counting vehicles and estimating their speeds across different lanes, we ensure precise detection for each lane. In object detection, tracking begins when the center point of the vehicle enters the starting line of the ROI range and continues until the vehicle leaves the endpoint line of the ROI range. The recorded positions and times of entry and exit from the ROI are used to estimate the vehicle's speed We validate the proposed method using the BrnoCompSpeed dataset, selecting videos with fast, medium, and slow vehicle speeds. The validation is performed using metrics such as average speed, standard deviation, variance, and accuracy, successfully demonstrating the accuracy and effectiveness of the proposed method. Furthermore, we compare our method with other approaches using metrics like mean error, median error, mean precision, mean recall, and processing speed to illustrate its strengths and weaknesses in terms of speed and accuracy. Finally, the proposed method is validated using real-time highway images provided by the Ministry of Transportation and Communications, demonstrating its usability and effectiveness.

摘要...............i
Abstract..........ii
誌謝..............iv
目錄...............v
表目錄.............vii
圖目錄.............viii
第一章 緒論.........1
1.1 研究背景........1
1.2 文獻回顧........1
1.3 研究動機........5
1.4 貢獻...........5
1.5 論文結構........5
第二章 車輛速度估算.......6
2.1 YOLO v5.............6
2.2 Region of Interest (ROI).......8
2.2.1車輛位置............8
2.2.2 估算車輛速度.......9
2.3 CCTV................10
2.4 BrnoCompSpeed.......10
第三章 實驗結果..........15
3.1實驗設備及處理平台....15
3.2 實驗結果............16
3.2.1車流量計算.........18
3.2.2估算車速...........21
3.3 BrnoCompSpeed數據集............23
3.3.1與現有架構比較......34
3.3.2比對有無遮罩........37
第四章 結論與未來展望.....39
4.1 結論................39
4.2 未來展望............40
參考文獻................41
Extended Abstract.......44
Chapter 1 Introduction............46
Chapter 2 Vehicle Speed Estimation.........46
Chapter 3 Experimental Results.......47
Chapter 4 Conclusion and Future Work.......48

[1]智慧運輸系統中長期發展規劃,交通部, 2019。
[2]國道即時路況, https://1968.freeway.gov.tw/
[3]Tsai, C.-M., Shih, F. Y., & Hsieh, J.-W. (2022). “Real-Time Vehicle Counting by Deep-Learning Networks.” https://doi.org/10.1109/icmlc56445.2022.9941299
[4]K. A. D. Devindu Dharmadasa, Goutam Kumar Sahoo, Santos Kumar Das, & Singh, P. (2023).“Video-based road accident detection on highways: A less complex YOLOv5 approach.”https://doi.org/10.1109/ic2e357697.2023.10262716
[5]Gupta, M., Kumar, R., Gupta, M., Kumar, M., Obaid, A. J., & Ved, C. (2024). “ YOLO-Based Vehicle Detection and Counting for Traffic Control on Highway. ” https://doi.org/10.1109/incacct61598.2024.10550995
[6]Pan, S.-H., & Wang, S.-C. (2021). “Identifying Vehicles Dynamically on Freeway CCTV Images through the YOLO Deep Learning Model.” Sensors and Materials, 33(5), 1517. https://doi.org/10.18494/sam.2021.3236
[7]高速公路1968國道壅塞排行, https://1968.freeway.gov.tw/tf_ranking
[8]Ahmed, S. et al. (2024).“Video based vehicle detection and tracking using image processing and deep learning.” 2024 3rd International Conference on Advancement in Electrical and Electronic Engineering (ICAEEE) [Preprint]. doi:10.1109/icaeee62219.2024.10561808.
[9]G Renuka Devi, S. Athiyan, k. Ilan Thamizh, & V. Hema Varthini. (2023). “ Speed and Distance Detection for Highway Traffic using Python.” https://doi.org/10.1109/icscan58655.2023.10395447
[10]林衍百，「利用高公局隧道內外即時影像估測行車速度」，碩士論文，國立交通大學數據科學與工程研究所，2020。
[11]Fernández Llorca, D., Hernández Martínez, A., & García Daza, I. (2021). “Vision‐based vehicle speed estimation: A survey.” IET Intelligent Transport Systems. https://doi.org/10.1049/itr2.12079
[12]UTFPR Datasets. (2018). Utfpr.edu.br. http://labic.utfpr.edu.br/datasets.html
[13]JakubSochor. (2016). GitHub - JakubSochor/BrnoCompSpeed: Code for BrnoCompSpeed dataset. GitHub. https://github.com/JakubSochor/BrnoCompSpeed
[14]Sonth, A., Settibhaktini, H. and Jahagirdar, A. (2019) “Vehicle speed determination and license plate localization from monocular video streams, ” Proceedings of 3rd International Conference on Computer Vision and Image Processing, pp. 267–277. doi:10.1007/978-981-32-9088-4_23.
[15]Luvizon, D.C., Nassu, B.T. and Minetto, R. (2016) “A video-based system for vehicle speed measurement in urban roadways, ” IEEE Transactions on Intelligent Transportation Systems, pp. 1–12. doi:10.1109/tits.2016.2606369.
[16]Luvizon, D.C., Nassu, B.T. and Minetto, R. (2014) “Vehicle speed estimation by license plate detection and tracking,” 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) [Preprint]. doi:10.1109/icassp.2014.6854869.
[17]Mini, T.V. and Vijayakumar, V. (2019).“Speed estimation and detection of moving vehicles based on probabilistic principal component analysis and New Digital Image Processing Approach,” EAI International Conference on Big Data Innovation for Sustainable Cognitive Computing, pp. 221–230. doi:10.1007/978-3-030-19562-5_22.
[18]Sochor, J., Juránek, R., Jakub Spanhel, Lukás Marsík, Široký, A., Herout, A., & Zemcik, P. (2019). “Comprehensive Data Set for Automatic Single Camera Visual Speed Measurement.” IEEE Transactions on Intelligent Transportation Systems, 20(5), 1633–1643. https://doi.org/10.1109/tits.2018.2825609
[19]Sochor, J., Juránek, R. and Herout, A. (2017) “Traffic surveillance camera calibration by 3D model bounding box alignment for accurate vehicle speed measurement,” Computer Vision and Image Understanding, 161, pp. 87–98. doi:10.1016/j.cviu.2017.05.015.
[20]Kocur, V. and Ftáčnik, M. (2020) “Detection of 3D bounding boxes of vehicles using perspective transformation for accurate speed measurement,” Machine Vision and Applications, 31(7–8). doi:10.1007/s00138-020-01117-x.
[21]Ultralytics (no date) Train Custom Data, GitHub. Available at: https://github.com/ultralytics/yolov5/wiki/Train-Custom-Data (Accessed: 03 July 2024).
[22]Sugiharto, H., Silviana, Y. and Nurpazrin, Y.S. (2023) “Unveiling document structures with Yolov5 layout detection,” arXiv.org. Available at: https://arxiv.org/abs/2309.17033 (Accessed: 03 July 2024).
[23]Cepeda-Negrete, J. et al. (2018) “Dark image enhancement using perceptual color transfer, ” IEEE Access, 6, pp. 14935–14945. doi:10.1109/access.2017.2763898.
[24]Chen, G. et al. (2019) “High-dynamic range, night vision, image-fusion algorithm based on a decomposition convolution neural network,” IEEE Access, 7, pp. 169762–169772. doi:10.1109/access.2019.2954912.
[25]Wang, K. and Liu, M.Z. (2020) “Object recognition at night scene based on DCGAN and faster R-CNN,” IEEE Access, 8, pp. 193168–193182. doi:10.1109/access.2020.3032981.
[26]Mo, X. et al. (2022) “Research on expressway traffic event detection at night based on mask-spynet,” IEEE Access, 10, pp. 69053–69062. doi:10.1109/access.2022.3178714.
[27]Zhang, W. et al. (2012) “Tracking and pairing vehicle headlight in night scenes, ” IEEE Transactions on Intelligent Transportation Systems, 13(1), pp. 140–153. doi:10.1109/tits.2011.2165338.