臺灣博碩士論文加值系統

高動態縮時攝影(Hyperlapse)是縮時攝影中的一種新興曝光技術，初期都是透過逐格拍攝單張影像來製作，過程相當費時。隨著錄影設備與高畫質技術的發展，現在許多使用者都普遍透過快轉預錄好的長時影片來製作，然而縮時的過程中會放大影片原本就存在的晃動，過度晃動的影片會使得觀賞者覺得頭暈和不舒服。
透過影片縮時的特性，縮時倍率決定了哪些畫面會被保留下來，但被挑選的畫面可能是晃動過大的畫面，進而造成常用的影片穩定方法無法有效處理的主要原因。在本篇研究中，我們提出了一種新的畫面選取(Frame Selection)方法來提升影片穩定程度。我們先找出畫面彼此之間匹配(Frame Matching)的資訊，並計算影像轉換成本(Transformation Cost)來評估各畫面的晃動幅度，之後使用畫面選取的方式保留轉換較小的畫面，取代原本由縮時倍率所選取的各畫面來達到減少影片晃動的效果，使得在我們使用影片穩定方法之前有更好的輸入影片。此外，我們的畫面選取方式可以有效地保持我們輸出影片的實際縮時倍率接近於我們預期的縮時倍率。
我們會與之前存在的演算法做比較，利用影片畫面之間X方向、Y方向與旋轉角度的轉換變化量程度大小來表示所提方法確實優於以往的演算法。

Hyperlapse is a new exposure technique in time-lapse photography. It needs to take pictures one by one and take a lot of time in the traditional way. Now we use camera to record long high quality videos and speed up them; however, if the video is shacking, the speed-up process will accentuate the unstable motion, resulting in a nauseating jumble.
Our goal is the hyperlapse videos stabilization. We present an algorithm for choosing the stable frame to replace the target speed-up frame, which is unstable. Our approach use the frame selection technique, we calculate the transformation cost to estimate if the frame is stable or not, by using frame matching for each video frame. Then we use our frame selection method based on transformation cost, this method will consider the transformation cost and target speed-up to make sure our hyperlapse videos is smoothing and the best match a desired target speed-up.
We will compare our result with other hyperlapse method by estimating the transformation of different directions for hyperlapse video, to prove our approach can make the hyperlapse video smoother.

第一章、緒論 1
1.1研究背景及動機 1
1.2 論文架構 3
第二章、文獻探討 4
2.1電影穩定(Cinema stabilization) 4
2.2第一人稱動態縮時攝影影片(First-person Hyper-lapse Videos) 5
2.3利用最佳畫面選取於即時建立動態縮時攝影(Real-Time Hyperlapse Creation via Optimal Frame Selection) 6
第三章、研究方法 9
3.1畫面匹配(Frame Matching) 10
3.1.1特徵點偵測(Feature Point Detection) 11
3.1.2特徵描述子與匹配(Feature Descriptor And Matching) 14
3.1.3射影變換矩陣(Homography Matrix) 15
3.1.4 RANSAC(RANdom SAmple Consensus) 16
3.2 成本估計(Cost Estimation) 19
3.3畫面選取(Frame Selection) 21
3.4影片穩定(Video Stabilization) 23
第四章、實驗結果及討論 26
4.1 態縮時攝影影片之結果 27
4.2 穩定成果比較 29
4.3 畫面裁切與轉換細節比較 37
第五章、結論與未來展望 41
參考文獻 42

[1]A. Karpenko, “The technology behind hyperlapse from instagram, ” http://instagramengineering.tumblr.com/post/95922900787/hyperlapse, August 13, 2014.
[2]N. Joshi, W. Kienzle, M. Toelle, M. Uyttendaele and M. Cohen, "Real-time hyperlapse creation via optimal frame selection," ACM Transactions on Graphics, vol. 34, no. 4, pp. 63:1-63:9, 2015.
[3]J. Kopf, M. Cohen and R. Szeliski, “First-person hyper-lapse videos,” ACM Transactions on Graphics, vol. 33, no. 4, pp. 1-10, July, 2014.
[4]H. Bay, A. Ess, T. Tuytelaars and L. Van Gool, “Speeded-Up Robust Features (SURF),” Computer Vision and Image Understanding, vol. 110, no. 3, pp. 346-359, 2008.
[5]M. Grundmann, V. Kwatra, and I. Essa, “ Auto-directed video stabilization with robust l1 optimal camera paths,” in Computer Vision and Pattern Recognition (CVPR), 2011 IEEE Conference on, pages 225–232, 2011.
[6]M. Calonder, V. Lepetit, C. Strecha, and P. Fua, “BRIEF: Binary robust independent elementary features,” in Proc. 11th Eur. Conf. Comput. Vis., May 2010, pp. 778–792.
[7]R. Hartley and A. Zisserman, Multiple view geometry in computer vision. Cambridge: Cambridge University Press, 2006.
[8]N. Livet, "Computing an homography between two views", http://iplimage.com/blog/geom-computing-homography-views/, July, 2012.
[9]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, vol. 24, no. 6, pp. 381-395, 1981.
[10]Y. Matsushita, E. Ofek, Weina Ge, Xiaoou Tang and Heung-Yeung Shum, “Full-frame video stabilization with motion inpainting,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 28, no. 7, pp. 1150-1163, 2006.
[11]F. Liu, M. Gleicher, H. Jin and A. Agarwala, “Content-preserving warps for 3D video stabilization,” ACM Transactions on Graphics, vol. 28, no. 3, p. 1, 2009.
[12]S. Baker, E. P. Bennett, S. B. Kang, and R. Szeliski, “Removing rolling shutter wobble,” in IEEE CVPR, 2010.
[13]E. Bennett and L. McMillan, “Computational time-lapse video,” in SIGGRAPH’07, July, 2007.
[14]C. Buehler, M. Bosse, and L. McMillan, “Non-metric image-based rendering for video stabilization,” in IEEE CVPR, 2001.
[15]J. Shi and C. Tomasi, “Good features to track,” in IEEE CVPR, 1994.
[16]Yair Poleg, Tavi Halperin, Chetan Arora, and Shmuel Peleg, “Egosampling: Fast-forward and stereo for egocentric videos,” in Computer Vision and Pattern Recognition (CVPR), 2015 IEEE Conference on, pp. 4768–4776, June, 2015.
[17]A. Goldstein and R. Fattal, "Video stabilization using epipolar geometry," ACM Transactions on Graphics, vol. 31, no. 5, pp. 1-10, 2012.
[18]P. Hargrave, “A tutorial introduction to Kalman filtering,” IEE Colloquium on Kalman Filters: Introduction, Applications and Future Developments, pp. 1-6, 1989.