臺灣博碩士論文加值系統

現今資訊科技蓬勃發展，電腦視覺技術經常應用於我們生活的周遭，而物件追蹤更是一項關鍵的技術，應用於自駕車、智慧行人追蹤和體育運動項目等領域。以籃球比賽中的球員為例，透過鏡頭追蹤球員在球場上的移動軌跡，可以對比賽進行詳細分析。

針對現有的一般追蹤方法（YOLOv7+StrongSORT），由於球員間的遮擋或重疊，常常會發生球員ID變換（ID Switch）且無法復原該球員原有的ID（Identifier）的情況。為了解決這一問題，我們提出了追蹤補償方法，該方法能在ID變換時匹配回先前的ID，從而提升球員追蹤的準確性。

在實驗結果中，我們選擇了在一般追蹤方法之下加入球員追蹤補償方法的架構（實驗組）以及僅使用一般追蹤方法的架構（對照組）進行比較。在MOTA（Multiple Object Tracking Accuracy）的數據上，對照組與實驗組的表現都高於90%。在評估球員ID變換時復原球員ID的整體ID變換復原率（ID Switch Recovery Rate）上，使用球員追蹤補償方法的實驗組得到了74%的整體ID變換復原率，而對照組只有48%。在整體追蹤準確度上，實驗組的IDF1（Identification F-Score）達到79%，而對照組則只有66%。從數據結果表明，使用球員追蹤補償方法後，整體ID變換復原率有明顯的提升，能夠減少球員ID在變換後無法復原的問題，從而使得在整體追蹤準確度上，IDF1得到顯著提升。

In the current era of rapidly advancing information technology, computer vision technology is frequently applied in our daily lives. Object tracking, a critical technology, is used in fields such as self-driving cars, intelligent pedestrian tracking, and sports. For example, in basketball games, cameras can track players' movements on the court, allowing for detailed analysis of the game.

Regarding existing general tracking methods (YOLOv7 + StrongSORT), player ID switching often occurs due to occlusion or overlap between players, and the original player ID (Identifier) cannot be recovered. To address this issue, we propose a tracking compensation method that can match the player’s previous ID during ID switching, thereby improving tracking accuracy.

According to experimental results, we compared a framework incorporating the player tracking compensation method (experimental group) with a framework using only the general tracking method (control group). Both frameworks achieved MOTA (Multiple Object Tracking Accuracy) scores above 90%. For the overall ID switch recovery rate, the experimental group using the player tracking compensation method achieved a 74% recovery rate, while the control group only achieved 48%. In terms of overall tracking accuracy, the experimental group's IDF1 (Identification F-Score) reached 79%, compared to 66% for the control group. The data results indicate that the use of the player tracking compensation method significantly improves the overall ID switch recovery rate, reduces the problem of player IDs not being restored after switching, and leads to a notable enhancement in overall tracking accuracy as reflected in the IDF1 score.

摘要 i
目錄 iii
圖目錄 v
表目錄 vii
第一章 緒論 1
1.1 研究動機 1
1.2 研究目的 2
1.3 論文架構 2
第二章 文獻探討 3
2.1 籃球比賽影像之球員追蹤挑戰 3
2.2 物件偵測 4
2.2.1 Two Stage Learning 4
2.2.2 One Stage Learning 7
2.3 多目標追蹤 9
2.3.1 運動特徵模型 10
2.3.2 結合運動與外觀特徵模型 12
第三章 研究方法 15
3.1 研究架構與流程 15
3.2球員偵測 16
3.3球員追蹤 17
3.4 球員追蹤補償方法 18
3.4.1 取得球員底部位置 19
3.4.2 Homograpy Transform 20
3.4.3 球員追蹤補償機制 21
第四章 實驗結果與討論 24
4.1 實驗環境 24
4.2 實驗影像資料集 25
4.3 實驗評估指標 26
4.3.1 MOTA 26
4.3.2 IDF1 27
4.3.3 ID變換復原率 27
4.4 實驗評估數據結果 28
4.5 實驗結果分析與討論 31
4.5.1 實驗組與對照組復原變換ID的分析與討論 31
4.5.2 追蹤補償方法未復原變換ID的分析與討論 34
第五章 結論與未來展望 36
5.1 結論 36
5.2 未來展望 37
參考文獻 38

[1]Bewley, A., Ge, Z., Ott, L., Ramos, F., Upcroft, B., “Simple Online and Realtime Tracking,” 2016 IEEE International Conference on Image Processing (ICIP), Phoenix, AZ, USA, 2016, pp. 3464-3468.
[2]Baker, S., Datta, A., Kanade, T., “Parameterizing homographies. Technical Report CMU-RI-TR06-11,” Carnegie Mellon University, Computer Science, thesis, 2014.
[3]Cai, Y., de Freitas, N., Little, J.J., “Robust Visual Tracking for Multiple Targets,” European Conference on Computer Vision, 2006.
[4]Cheshire, E., Hu, M.C., Chang, M.H., “Player Tracking and Analysis of Basketball Plays,” Leland Stanford Junior University, Computer Science, thesis, 2015.
[5]Du, Y., Zhao, Z., Song, Y., Zhao, Y., Su, F., Gong, T., Meng, H., “StrongSORT: Make DeepSORT Great Again,” in IEEE Transactions on Multimedia, vol. 25, pp. 8725-8737, 2023.
[6]Girshick, R., Donahue, J., Darrell, T., Malik, J., "Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation," 2014 IEEE Conference on Computer Vision and Pattern Recognition, Columbus, OH, USA, 2014, pp. 580-587.
[7]Girshick, R., “Fast R-CNN,” 2015 IEEE International Conference on Computer Vision (ICCV), Santiago, Chile, 2015, pp. 1440-1448.
[8]He, K., Gkioxari, G., Dollár, P., Girshick, R., “Mask R-CNN,” 2017 IEEE International Conference on Computer Vision (ICCV), Venice, Italy, 2017, pp. 2980-2988.
[9]Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.Y., Berg, A.C., “SSD: single shot multibox detector [C]”, Proceedings of the 14th European Conference on Computer Vision ECCV 2016, pp. 21-37, 2016.
[10]Lu, W.L., Ting, J.A., Little, J.J., Murphy, K.P., “Learning to Track and Identify Players from Broadcast Sports Videos,” in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 35, no. 7, pp. 1704-1716, July 2013.
[11]Parsons, S., Rogers, J., “Basketball Player Tracking and Automated Analysis” Leland Stanford Junior University, Computer Science, thesis, 2014.
[12]Redmon, J., Divvala, S., Girshick, R., Farhadi, A., “You Only Look Once: Unified, Real-Time Object Detection,” 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 2016, pp. 779-788.
[13]Ren, S., He, K., Girshick, R., Sun, J., “Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks,” in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 6, pp. 1137-1149, 1 June 2017.
[14]Rosebrock, A., “Simple object tracking with OpenCV,” pyimagesearch, https://pyimagesearch.com/2018/07/23/simple-object-tracking-with-opencv/, July 2018 (accessed June 2024).
[15]Tamir, M., et al., “SportVU,” statsperform, https://www.statsperform.com/team-performance/basketball/optical-tracking/, July 2005 (accessed June 2024).
[16]Wang, C.Y., Bochkovskiy, A., Liao, H.Y.M., “YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real-Time Object Detectors,” 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Vancouver, BC, Canada, 2023, pp. 7464-7475.
[17]Wen, N., Dalbo, V.J., Burgos, B., Pyne, D.B., Scanlan, A.T., “Power testing in basketball: Current practice and future recommendations,” The Journal of Strength & Conditioning Research, 2018.
[18]Wojke, N., Bewley, A., Paulus, D., “Simple online and realtime tracking with a deep association metric,” 2017 IEEE International Conference on Image Processing (ICIP), Beijing, China, 2017, pp. 3645-3649.
[19]Xing, J., Ai, H., Liu, L., Lao, S., “Multiple Player Tracking in Sports Video: A Dual-Mode Two-Way Bayesian Inference Approach With Progressive Observation Modeling,” in IEEE Transactions on Image Processing, vol. 20, no. 6, pp. 1652-1667, June 2011.
[20]Yoon, Y., Hwang, H., Choi, Y., Joo, M., Oh, H., Park, I., Lee, K.H., Hwang, J.H., “Analyzing Basketball Movements and Pass Relationships Using Realtime Object Tracking Techniques Based on Deep Learning,” in IEEE Access, vol. 7, pp. 56564-56576, 2019.
[21]Zhang, Y., Sun, P., Jiang, Y., Yu, D., Weng, F., Yuan, Z., Luo, P., Liu, W., Wang, X., “Bytetrack: Multi-object tracking by associating every detection box,” 2021, [online] CoRR vol. abs/2110.06864.
[22]Zhang, Y., Wang, C., Wang, X., Zeng, W., Liu, W., “FairMOT: On the fairness of detection and re-identification in multiple object tracking,” Int. J. Comput. Vis., vol. 129, no. 11, pp. 3069–3087, 2021.
[23]Zheng, L., Yang, Y., Hauptmann, A.G., “Person Re-identification: Past, Present and Future,” 2016, [online] ArXiv, abs/1610.02984.
[24]上班族歡樂籃球聯盟，https://www.youtube.com/@user-yj9tn6np9u。