隨著科技的進步和運動賽事轉播需求的增加，轉播的畫面往往會受限於單一攝影機的視野大小 (FOV, Field of View) 以及攝影機的架設位置，無法一次性完整呈現整個運動場域。為了改善此問題，本研究致力於開發出一個精確的影像拼接系統，能夠將從多個不同視角拍攝的影像無縫拼接成全景圖像，以擴展能觀看的體育場域視野範圍。我們的目標包括設計和實現一個能夠正確且可調整的影像拼接演算法，以滿足不同體育賽事和攝影機配置的需求。我們的作法如下，首先，我們會根據拍攝的運動場域建立一個虛擬場域平面，這個虛擬平面會參考現實中的運動場域規格。第二步，使用TrackNet偵測影像中球場的邊界點，通過虛擬平面邊線座標點的設定，精確地找到兩者間的對應點並將拍攝的影像投影到虛擬平面上。接著，我們會使用GraphCut演算法，找到投影到虛擬平面後相鄰的影像間的接縫，並在兩兩之間保留40像素的重疊區域，以供後續相鄰影像間的混合處理。最後，使用曝光錯誤補償及多頻段混合技術處理重疊區域，實現影像間的平滑過渡，確保拼接後圖像的自然流暢和視覺質量。這項研究成果將顯著提升運動賽事轉播的觀賞體驗，擴展觀眾的視覺感知範圍，使他們能夠更全面地體驗和欣賞賽事場地的動態和廣度。儘管本論文所提出的拼接方法能有效地在多種不同的運動場域中實現高質量的拼接結果，但仍然面臨很多挑戰，如：無法實現即時影像的拼接、動態物體在移動過程經過接縫時可能會產生拼接錯誤等，期望在未來能透過加速演算法、來嘗試解決，使本論文的研究和發展更加完善，讓其在發揮更大的作用。

With the advancement of technology and the increasing demand for sports broadcasting, the coverage is often limited by the field of view (FOV) of a single camera and its positioning, preventing a comprehensive presentation of the entire sports venue. To address this issue, our research aims to develop an accurate image stitching system capable of seamlessly merging images captured from multiple perspectives into a panoramic view, thereby expanding the observable field of sports venues.Our objectives include designing and implementing a robust and adjustable image stitching algorithm to meet the needs of different sports events and camera setups. Our approach is as follows: first, we construct a virtual field plane based on the specifications of the real sports venue. Second, we use TrackNet to detect the boundary points of the court in the images, accurately locating corresponding points between the images and the virtual plane through the setting of boundary coordinates on the virtual plane, and projecting the captured images onto this virtual plane. Next, we employ the GraphCut algorithm to identify seams between adjacent images projected onto the virtual plane, maintaining an 40-pixel overlap between each pair of adjacent images to facilitate subsequent blending processes. Finally, we apply exposure compensation and multi-band blending techniques to the overlapping areas to achieve smooth transitions between images, ensuring the natural flow and visual quality of the stitched image.This research outcome will significantly enhance the viewing experience of sports broadcasts by expanding the visual perception range of the audience, allowing them to comprehensively experience and appreciate the dynamics and breadth of the sports venue. Although the proposed stitching method effectively achieves high-quality stitching results in various sports venues, it still faces several challenges, such as the inability to perform real-time image stitching and potential stitching errors when dynamic objects move across seams. We hope to address these issues in the future by accelerating the algorithm, thereby improving and expanding the impact of our research.

[1] H. Bay, T. Tuytelaars, and L.V. Gool, “Surf: Speeded up robust features, ” Computer Vision ECCV, pp. 404-417, 2006.
[2] G. Bradski, and A. Kaehler, “OpenCV,” Dr. Dobb’s Journal of Software Tools, 2000.
[3] P.J. Burt and E.H. Adelson, “A Multiresolution Spline With Application to Image Mosaics, ” Acm transactions on graphics (ToG), pp. 217-236, 1983.
[4] M. Brown and D.G. Lowe, “Automatic panoramic image stitching using invariant features, ” International journal of computer vision, pp. 59–73, 2007.
[5] T. Chen, “Real-Time Streaming Synchronization of Multiple IP Cameras Based on RTSP Protocol and Gigapixel Image Viewing,” 2024.
[6] O. Chum, T. Pajdla, and P. Sturm, “The geometric error for homographies,” Computer Vision and Image Understanding, pp. 86–102, 2005.
[7] D. DeTone, T. Malisiewicz, and A. Rabinovich, “Deep image homography estimation,” arXiv, 2016.
[8] M. Fischler and R. Bolles, “Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,” Communications of the ACM, pp. 381-395, 1981.
[9] J.D. Foley,“Computer graphics: principles and practice,” Addison-Wesley Professional, 1996.
[10] M. Fu, H. Liang, C. Zhu, Z. Dong, R. Sun and Y. Yue,“Image Stitching Techniques Applied to Plane or 3-D Models: A Review,” in IEEE Sensors Council (iSensors), pp. 8060-8079, 2023.
[11] J. Gao, S.J. Kim and M.S. Brown, “Constructing Image Panoramas using Dual-Homography Warping,” in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 49-56, 2011.
[12] J. Hui, “CUDA Tutorial”, [Online]. Available: https://jhui.github.io/2017/03/06/CUDA/.
[13] Ketkar, Nikhil, “Introduction to PyTorch,” pp. 195–208, 2017.
[14] V. Kwatra, A. Schodl, I. Essa, G. Turk and A. Bobick, “Graphcut Textures: Image and Video Synthesis using Graph Cuts, ” Acm transactions on graphics (ToG), pp. 277-286, 2003.
[15] W.Y. Lin, S. Liu, Y. Matsushita, T.T. Ng, and L.F. Cheong, “Smoothly varying affine stitching,” in Computer Vision and Pattern Recognition (CVPR), pp. 345–352, 2011.
[16] D.G. Lowe, “Distinctive image features from scale-invariant keypoints,” International Journal of Computer Vision, pp. 91-110, 2004.
[17] M. Muja and D.G. Lowe, “Fast approximate nearest neighbors with automatic algorithm configuration,” VISAPP, pp.331-340, 2009.
[18] L. Nie, C. Lin, K. Liao, S. Liu and Y. Zhao, “Learning Thin-Plate Spline Motion and Seamless Composition for Parallax-Tolerant Unsupervised Deep Image Stitching,” arXiv, pp.1-19, 2023.
[19] L. Nie, C. Lin, K. Liao, S. Liu and Y. Zhao, “Deep Rectangling for Image Stitching: A Learning Baseline, ” in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 5740-5748, 2022.
[20] A. Paszke, S. Gross, S. Chintala and G. Chanan, “Pytorch CUDA Document,” [Online] Available: https://pytorch.org/docs/stable/notes/cuda.html.
[21] M. Uyttendaele, A. Eden and R. Szeliski,“Eliminating Ghosting and Exposure Artifacts in Image Mosaics, ” in IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), pp.509-516, 2001.
[22] K. Sergey,“Tennis analysis using deep learning and machine learning,” [Online] Available: https://medium.com/@kosolapov.aetp/tennis-analysis-using-deep-learning-and-machine-learning-a5a74db7e2ee.
[23] R. Szeliski, “Image alignment and stitching: A tutorial, ” Foundations and Trends in Computer Graphics and Vision, pp.1-104, 2007.
[24] H. Wu, S. Zheng, J. Zhang and K Huang, “GP-GAN: Towards realistic high-resolution image blending, ” Proceedings of the 27th ACM international conference on multimedia, 2019.
[25] J. Zaragoza, T.J. Chin, M.S. Brown and D. Suter, “As-Projective-As-Possible Image Stitching with Moving DLT, ” in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2339-2346, 2013.
[26] F. Zhang and F. Liu, “Parallax-Tolerant Image Stitching, ” in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4321-4328, 2014.
[27] J. Zhang, C. Wang, S. Liu, L. Jia, N. Ye, J. Wang, J. Zhou, and J. Sun, “Content-Aware Unsupervised Deep Homography Estimation,” Computer Vision ECCV, pp. 1-18, 2020.
[28] L. Zhang, T. Wen and J. Shi, “Deep image blending, ” Proceedings of the IEEE/CVF winter conference on applications of computer vision, 2020.