臺灣博碩士論文加值系統

隨著經濟的成長台灣汽車數量因此也隨之增加，而交通事故與違規事件的比率亦顯著成長，為維護交通道路的安全與減輕執法人員的負擔，有效應用科技執法AI人工智慧(Artificial intelligence)是目前熱門的話題。所以本研究提出透過在道路監視器畫面上，應用電腦視覺和機器學習的方式來自動判別車輛是否做出交通違規的舉動。本研究提出一個以軌跡追蹤及非監督式資料分群為基礎的偵測方法，軌跡追蹤採用了DeepSORT(Deep Simple Online and Realtime Tracking)模型來追蹤車輛並獲取軌跡，此模型是目前較常被引用的模型，但該模型較易出現軌跡追蹤異常的問題，在於非監督式資料分群部分，常見文獻中採用軌跡的起點、終點以及出現時間等相關資訊作為分群的向量資料，這樣的分群方式較無法精確有效的找出異常軌跡。本研究為使在判別上更具準確性，提出兩項改進措施，首先為了確保資料的完整性，由DeepSORT產生的軌跡會先去除極端的軌跡長度和轉彎角度，另外，軌跡內密度較大的座標點群會只留下一個座標點其他座標點則被視為冗餘資料刪除，接著依照本研究的相似度評估方式採用階層式分群法與DBSCAN(Density-based spatial clustering of applications with noise)進行分群，分好的群組會被用來作為新軌跡是否為異常軌跡的依據。為了比較結果，本研究使用3個自行攝影的資料集與6個公開資料集做為比較標準，當使用以提出的方法作為相似度評估方法的DBSCAN時能夠提升分群的效果，本研究建立的模式期望在未來提供更多電腦視覺和機器學習的技術發展方向及其應用，並藉此模式使其在判別上更具準確性。

With the economic growth, the number of cars in Taiwan has also increased. The ratio of traffic accidents and violations has also increased significantly. In order to maintain the safety of traffic roads and reduce the burden on law enforcement officers. AI (Artificial intelligence) is a effectively and useful tool. Therefore, this study proposes to automatically determine whether the vehicle has committed traffic violations by applying computer vision and machine learning on the road monitor screen. This study proposes a detection method based on trajectory tracking and unsupervised learning, to fomd abnormal trajectory. DeepSORT (Deep Simple Online and Realtime Tracking) is the model to track vehicles and obtain trajectories. But this model has so many problem. In common literature, relevant information use the starting point and ending point to build a time of the trajectory are used as the vector data for grouping. Such a grouping method is not accurate. Effectively find out the abnormal trajectory. In order to make the discrimination more accurate, this study puts forward two improvement measures. First, in order to the integrity of the data, the trajectory generated by DeepSORT will first remove the extreme trajectory length and turning angle. Only one coordinate point will be left when the points density too large. Then, base on the proposed similarity evaluation method, the hierarchical clustering method and DBSCAN (Density-based spatial clustering of applications with noise) will be used for grouping. The divided groups will be used as the basis for whether the new trajectory is an abnormal trajectory. In order to compare the results, this study uses 3 self-photographed data sets and 6 public data sets as comparison standards. When using the proposed method as the DBSCAN base on similarity evaluation method can improve the effect of grouping. The model expects to provide more technical development directions and applications of computer vision and machine learning in the future.

摘要 ...i
Abstract ...ii
誌謝 ...iv
目錄 ...v
表目錄 ...vii
圖目錄 ...ix
符號說明 ...xiv
第一章 緒論 ...1
1.1 研究背景與動機 ...1
1.2 研究目的 ...2
1.3 研究架構 ...2
第二章 文獻探討 ...3
2.1 異常軌跡偵測之相關文獻 ...3
2.1.1 監督式作法異常軌跡偵測 ...3
2.1.2 非監督式異常軌跡偵測 ...8
2.2 多目標追蹤之相關研究 ...9
2.3 分群法相關研究 ...12
2.3.1 K-means ...12
2.3.2 階層式分群法（hierarchical clustering） ...12
2.3.3 DBSCAN(Density Based Clustering) ...13
2.4 軌跡相似度評估之相關研究 ...14
2.4.1 豪斯多夫距離(Hausdorff distance) ...14
2.4.2 最長共同子序列 ...14
2.4.3 Piecewise Space-Time Constraints ...15
第三章 研究內容與方法 ...17
3.1 系統架構與流程 ...17
3.2 軌跡資料蒐集 ...17
3.3 資料預處理 ...18
3.3.1 去除冗餘座標資訊 ...19
3.3.2 去除角度異常軌跡 ...20
3.3.3 去除軌跡內座標資訊太少軌跡 ...20
3.3.4 去除追蹤失敗軌跡 ...21
3.4 軌跡分群建立模型 ...22
3.4.1 軌跡相似度評估 ...22
3.4.2 軌跡類別相似度評估 ...26
3.5 利用模型偵測異常軌跡 ...27
第四章 實驗結果 ...30
4.1 採用的資料集 ...30
4.2 分群效能的測量 ...30
4.3 實驗參數與實驗環境 ...33
4.4 軌跡分群 ...48
4.4.1. 正常軌跡分群結果 ...48
4.4.2. 異常軌跡偵測結果 ...85
4.4.3. 實驗結果討論與總結 ...105
第五章 結論 ...118
參考文獻 ...119
Extended Abstract ...122

[1]交通部交通安全督導委員會，道安資訊查詢網。檢自https://roadsafety.tw/Dashboard/Custom? type=統計快覽, (2022年5月13日)。
[2]Y. Ji, L. Wang, W. Wu, H. Shao and Y. Feng, 2020, "A Method for LSTM-Based Trajectory Modeling and Abnormal Trajectory Detection," IEEE Access, vol. 8, pp. 104063-104073, May.
[3]Cheng, Y., Wu, B., Song, L., Shi, C, 2019. "Spatial-Temporal Recurrent Neural Network for Anomalous Trajectories Detection.", Advanced Data Mining and Applications, pp 565–578, November.
[4]Cheng, Bin, Qian, Shiyou, Cao, Jian, Xue, Guangtao, Yu, Jiadi, Zhu, Yanmin, Li, Minglu, Zhang, Tao, 2019, "STL: Online Detection of Taxi Trajectory Anomaly Based on Spatial-Temporal Laws", Database Systems for Advanced Applications, pp 764–779, April.
[5]Song L., Wang R., Xiao D., Han X., Cai Y., Shi C, 2018, "Anomalous Trajectory Detection Using Recurrent Neural Network", Advanced Data Mining and Applications. ADMA 2018, pp. 263–277, December.
[6]X. Zhao, Y. Rao, J. Cai and W. Ma, 2020, "Abnormal Trajectory Detection Based on a Sparse Subgraph", IEEE Access, vol. 8, pp. 29987-30000, February.
[7]K. K. Santhosh, D. P. Dogra, P. P. Roy and A. Mitra, 2021, "Vehicular Trajectory Classification and Traffic Anomaly Detection in Videos Using a Hybrid CNN-VAE Architecture," IEEE Transactions on Intelligent Transportation Systems, vol. 23, pp. 1-12, September.
[8]P. R. Roy and G. Bilodeau, 2019, "Adversarially Learned Abnormal Trajectory Classifier," 2019 16th Conference on Computer and Robot Vision (CRV), pp. 65-72, May.
[9]Doan Trung, Kim Sunwoong, Vo Le Cuong, Lee Hyuk-Jae, 2016, "Anomalous Trajectory Detection in Surveillance Systems Using Pedestrian and Surrounding Information", IEIE Transactions on Smart Processing and Computing, vol. 5, pp. 256-266, August.
[10]K. K. Santhosh, D. P. Dogra, P. P. Roy and B. B. Chaudhuri, 2021, "Trajectory-Based Scene Understanding Using Dirichlet Process Mixture Model, " IEEE Transactions on Cybernetics, vol. 51, pp. 4148-4161, Aug.
[11]Jiang Haobin, Cai Yingfeng, Wang Hai, Chen Xiaobo, 2015, "Trajectory-based anomalous behaviour detection for intelligent traffic surveillance", IET Intelligent Transport Systems, vol. 9, pp. 810-816, April.
[12]Sekh, A.A., Dogra, D.P., Kar, S. et al, 2020, “Video trajectory analysis using unsupervised clustering and multi-criteria ranking.”, Soft Comput, vol. 24, pp 16643–16654, May.
[13]Petitjean F, Forestier G, Webb G, Nicholson AE, Chen Y, Keogh E et al, 2014, “Dynamic time warping averaging of time series allows faster and more accurate classification.”, IEEE international conference on data mining (ICDM), pp 470–479, January
[14]Murdock, D., & Martinez del Rincon, J., 2016, “Abnormal Pedestrian Trajectory analysis based on arbitrary-length clustering”, Irish Machine Vision & Image Processing Conference Proceedings 2016, pp. 17-24, August.
[15]Z. Sun, J. Chen, L. Chao, W. Ruan and M. Mukherjee, 2021, "A Survey of Multiple Pedestrian Tracking Based on Tracking-by-Detection Framework," IEEE Transactions on Circuits and Systems for Video Technology, vol. 31, pp. 1819-1833, May.
[16]G. Brasó and L. Leal-Taixé, 2020, "Learning a Neural Solver for Multiple Object Tracking," 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 6246-6256, April.
[17]W. Ren, X. Wang, J. Tian, Y. Tang and A. B. Chan, 2020, "Tracking-by-Counting: Using Network Flows on Crowd Density Maps for Tracking Multiple Targets", IEEE Transactions on Image Processing, vol. 30, pp. 1439-1452, December.
[18]A. Bewley, Z. Ge, L. Ott, F. Ramos and B. Upcroft, 2016, "Simple online and realtime tracking", IEEE International Conference on Image Processing (ICIP), pp. 3464-3468, August.
[19]N. Wojke, A. Bewley and D. Paulus, 2017, "Simple online and realtime tracking with a deep association metric", 2017 IEEE International Conference on Image Processing (ICIP), pp. 3645-3649, March.
[20]R. Laxhammar and G. Falkman, 2011, "Sequential Conformal Anomaly Detection in trajectories based on Hausdorff distance",14th International Conference on Information Fusion, pp. 1-8, July.
[21]M. Y. Choong, L. Angeline, R. K. Yin Chin, K. Beng Yeo and K. T. Kin Teo, 2017, "Modeling of vehicle trajectory clustering based on LCSS for traffic pattern extraction", IEEE 2nd International Conference on Automatic Control and Intelligent Systems (I2CACIS), pp. 74-79, October.
[22]C. Liu, J. Wang, A. Liu, Y. Cai and B. Ai, 2020, "An Asynchronous Trajectory Matching Method Based on Piecewise Space-Time Constraints," in IEEE Access, vol. 8, pp. 224712-224728, December.
[23]T. Hospedales, S. Gong and T. Xiang, 2012, “Video Behaviour Mining Using A Dynamic Topic Model”, International Journal of Computer Vision (IJCV), vol. 98, pp. 303-323, July.

[24]T. Hospedales, S. Gong and T. Xiang, 2012, “Video Behaviour Mining Using A Dynamic Topic Model”, International Journal of Computer Vision (IJCV), vol. 98, pp. 303-323, July.
[25]J. Li, S. Gong and T. Xiang, 2012, “Learning Behavioural Context”, International Journal of Computer Vision (IJCV), vol. 97, pp. 276-304, May.
[26]X. Wang, K. T. Ma, G. Ng and E, 2008, “Grimson, Trajectory Analysis and Semantic Region Modeling Using A Nonparametric Bayesian Model”, in Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), pp. 1-8, June.
[27]X. Wang, K. Tieu and E. Grimson, 2010, “Correspondence‐Free Activity Analysis and Scene Modeling in Multiple Camera Views”, IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), vol. 32, pp. 56-71, October.
[28]Steffen Busch, Christian Koetsier, Jeldrik Axmann, 2022, “Dataset: LUMPI: The Leibniz University Multi-Perspective Intersection Dataset”, pp. 1127-1134, June.