ROBUST VIDEO STABILIZATION BASED ON PARTICLE FILTER TRACKING OF PROJECTED CAMERA MOTION IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, 2009 Keywords: Bootstrap filtering, motion analysis, Particle filtering, video stabilization P48981023 冷玉琦 Yu-Chi Leng 2011/01/16 ABSTRACT We propose a novel technique for video stabilization based on the particle filtering framework. Extend the traditional use of particle filters in object tracking to tracking of the projected affine model of the camera motions. The correspondence between scale-invariant feature transform points is used to obtain a crude estimate of the projected camera motion. INTRODUCTION Video cameras mounted on handheld devices and mobile platforms have become increasingly popular in the consumer market over the past few years due to a dramatic decrease in the cost of such devices. Stabilization methods exploit the fact that camera motion causes the affine transform of the frames, which can be inverted to obtain stable frames. THEORETICAL FOUNDATIONS A. Camera Model 攝影機移動前後向量關係 [ x1 y1 z1 ] R 3 x 3 [ x 0 [u 0 v0 ] [u1 v1 ] T T z0 u1 R 11 s v1 R 21 T z1 [ x0 y0 [ x1 y1 z1 ] 1 R11 R12 2 T z0 ] R 12 u 0 t x v t R 22 0 y R11 R 21 R12 R 22 z 0 ] T3 x 1 y0 2 T T 6個自由度! 1 R 21 R 22 0 2 2 THEORETICAL FOUNDATIONS B. Particle Filtering Estimation 狀態向量: x [ s , R , R , R , t , t ] 事後機率密度函數:p ( x z ) T k k 11 k 12 k k N 21 k xk yk 1: k p ( x k z 1:k ) w k ( x k x k ) i i i 1 w k p ( x k z 1:k ) q ( x k z 1:k ) i 估算當前狀態: xˆ k E x k x k p ( x k z 1:k ) d x k x k w k ( x k x k )d x k w k x k N i 1 重要性密度: x i k ~ q(x k , 1 ) i i N i 1 i i THEORETICAL FOUNDATIONS 誤差向量 k xˆ k x k ek xk xk 協方差矩陣 Cov ( k , k ) Cov ( k , k ) Cov ( e k , e k ) 1 ( 1 2 ) c k N Cov ( k , k ) Cov ( e k , e k ) 2 1 N 2 1j 2 j c k 2 j N max 2 j 1 ... M 2 c k 2 1j 2j 2j Nk 2 2 O VIDEO STABILIZATION A. Importance Density Using Scale-Invariant Features 影像間的特徵點將使用SIFT獲得。 使用特徵追蹤來得到中值向量: x [ s , R , R , R , t , t ] k u k .. vk u k 1 s .. .. s k R11 k 1 s k R 21 k 1 tx v k 1 .. k x k ~ qG (x k , 1 ) i k 11 k 12 k 21 k xk yk s k R12 k s k R 22 k t y k T 1 i 1 i exp x x x x k 1 k k 6 2 k ( 2 ) 1 1 有效的減少粒子數至30個並有同等或超越300個粒子 所得的品質。 T VIDEO STABILIZATION B. Particle Filtering for Global Motion Estimation Between Successive Frames -權重值依粒子有多靠近真實的狀態來給定。 -選擇均方差(mean square error, MSE)及特徵距離做為 相似度的兩個量測量。 PMSE i Mi exp 2 2 M 2 1 2 σ M i wk i P feature i D exp i 2 2 σ F 2 F 1 2 i PMSE P feature i i i 1 PMSE P feature N VIDEO STABILIZATION 利用離散權重近似真實狀態,得到仿射動態參數: Rˆ 11 k xˆ k [ sˆ k Rˆ 12 k Rˆ 21 k tˆxk T tˆyk ] 尺度因子、旋轉矩陣和位移與參考影像的關係: s k s k 1 sˆ A Tk A A Rˆ 11 k sˆ k ˆ R 21 k A Rk Rˆ 11 k R k 1 ˆ R 21 k A tˆx Rˆ 12 k A T k 1 ˆ ˆ R 22 k t y k k Rˆ 12 k Rˆ 22 k VIDEO STABILIZATION C. Intentional Motion Estimation and Motion Compensation 沿著x方向的平移和平移速度 將可表示為: Tx k 1 v Tx k 0 1 Tx k 1 0 Tx v 1 k 1 n 補償非預期動態: D u k u s s k D A 1 A D A R k R k T k T k vs sk vk [ u k ,v k ] 不穩定影像像素點 [ u s ,v s ] 補償後穩定影像像素點 VIDEO STABILIZATION Algorithm: EXPERIMENTAL RESULTS EXPERIMENTAL RESULTS EXPERIMENTAL RESULTS EXPERIMENTAL RESULTS EXPERIMENTAL RESULTS EXPERIMENTAL RESULTS CONCLUSION In this paper, we presented a novel approach for robust video stabilization based on particle filter estimation of projected camera motion. An efficient implementation of particle filters for global motion estimation has been proposed based on carefully designed importance sampling. We demonstrated experimentally that the proposed particle filtering scheme can be used to obtain an efficient and accurate motion estimation in video sequences.