基於SURF特徵的全域運動影片之穩定性改善

Improving Video Stabilization Based on SURF Feature for Global Motion

宋俊宏(Chun-Hung Sung)

指導教授 ： 林春宏

國立臺中科技大學/資訊與流通學院/資訊工程系碩士班/碩士(2017年)

摘要

由於資訊科技技術的進步，影像穩定技術相關產品在市面上規格有些差異，如有電子影像穩定(EIS)、光學影像穩定(OIS)、數字影像穩定(DIS)等技術，都會造成拍攝的影像有差異。若是加上一些不可避免的因素，如環境或人為等因素，而造成影像模糊不清，因此我們就必須去探討如何克服將影像穩定的問題。
為了要解決攝影機振動所造成的視訊振動，本文針對影像內容進行分析，得出造成影像振動因素主要分為兩種:物件的運動與攝影機的運動。第一種攝影機內物件運動為局部性的變化，第二種攝影機振動為全域性的變化。
本文採用全域運動估測(global motion estimation)方法來進行影像的矯正。首先使用SURF找出影片的SURF特徵點，接著再使用L-K光流法找出區域運動(local motion)，並且利用區域運動(local motion)對SURF特徵點進行篩選，進而取得全域運動(global motion,GM)特徵點，再使用RANSAC特徵點匹配，以便找到兩張影像間的轉換矩陣參數。接著估測視訊頁(frame)的振動行為，本研究將從轉換矩陣參數找到像素間位移的對應關係。然後再針對像素間的位移進行影像全域性矯正，最後將校正資料經卡爾曼濾波(kalman filter)得到預測的矯正值，以使得視訊頁(frame)內的運動達到平穩。
為了驗證本研究方法的實用性，本文針對SURF特徵點與全GM特徵點之穩定效能進行實驗比對。透過實驗結果得知本文所提出之方法能有效地將區域運動特徵點篩選出來，全域運動特徵點之穩定性也隨著區域運動特徵點之去除變得更加穩定，藉由穩定前後影像與PSNR值都可以明顯看出，本文所提出的方法在影像穩定上能有效的處理影像震動所帶來的位移。

關鍵字

SURF ； 光流法 ； 轉換矩陣 ； 卡爾曼濾波 ； 全域運動 ； 區域運動

並列摘要

The significant growth of current information technology (IT) has given rise to a number of differences in the specifications of video stabilization technology and related products. For example, images will differ when using electronic image stabilization (EIS), optical image stabilization (OIS), digital image stabilization (DIS) or other similar technologies. Unavoidable factors, such as environmental or human factors, which cause image distortion, must also be taken into account. Thus, the problem of image stabilization in current imaging technology is a significant industry challenge.
In order to address video vibration caused by the physical movement of the camera, this paper proposes image content analysis, in which image vibration factors are divided into two types: the movement of objects and the movement of the camera. Camera vibration can then be divided in two categories: local motion, and global motion.
In this paper, the global motion (GM) estimation method is used for image stabilization. First, Speeded Up Robust Features (SURF) is used to find the SURF feature points of the video, and then the L-K optical flow method is used to identify local motion in the video. The SURF feature points are then screened by local motion to obtain the GM feature points, and Random sample consensus (RANSAC) feature points are then used to find the homograph matrix parameters between the two images. After this, the vibration behavior of the video is estimated. This study identifies the correspondence between the displacements of the pixels from the conversion matrix parameters. The image is then interpolated for the global correction of the image, and the correction data by the Kalman filter is finally used to obtain the predicted correction value, so as to achieve a smooth frame motion.
In order to verify the practicability of the proposed method, this paper compares the SURF feature points with the global motion feature points. From the experiment results, it is found that the proposed method is able to effectively select the regional motion feature points, and the stability of the global moving feature points becomes enhanced with the removal of the regional motion feature points. Stabilizing the image before and after the peak signal to noise ratio (PSNR), clearly shows that the proposed method can effectively deal with the displacement caused by image vibration in image stabilization.

並列關鍵字

SURF ； optical flow ； homograph matrix ； Kalman filter ； global motion ； local motion

參考文獻

[7] A. Censi, A. Fusiello, V. Roberto, “Image stabilization by features tracking,” in: International Conference on Image Analysis and Processing, pp.665–667, 1999.

[1] C.H. Chen, Y.L. Kuo, “Real-time Video stabilization based on motion compensation,” Innovative Computing, Information and Control (ICICIC), pp.1495–1498,2009.

[2] D. Lowe, “Distinctive image features from scale invariant keypoints,” International Journal of Computer Vision 60, pp.91–110,2004.

[5] Y. Junlan, D. Schonfeld, M. Mohamed, “Robust video stabilization based on particle filter tracking of projected camera motion,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 19, pp.945–954, 2009.

[6] J. Chang, W. Hu, M. Cheng, M. Chang, “Digital image translational and rotational motion stabilization using optical flow technique,” IEEE Transactions on Consumer Electronics, vol. 48, pp.108–115, 2002.