Efficient camera motion characterization for MPEG video indexing

Camera Motion Analysis in On-line MPEG Sequences

“…The most popular models in camera motion estimation from MPEG motion vectors are the four parameter geometric model [3] and the six parameters affine model [2]. Both can represent, under certain geometric assumptions, the optical flow resulting from the camera pan/track, tilt/boom, dolly/zoom or roll.…”

“…The most common strategy to estimate camera motion via fitting to a motion model usually consists of three steps: the fitting procedure itself (in which quite often only P frames are considered), the analysis of the resulting parameters to decide on the camera motion type, and a post-filtering stage to discard too short motions (typical threshold values are 0.5 seconds [2] or 10 frames [1]). …”

“…The problem is tackled through the fitting of the motion field to a camera motion model [2] [3], that is to estimate the set of parameters that better describe the transformation of the pixels (or blocks) between two consecutive frames. The value of these parameters is afterwards analyzed to obtain the camera motion patterns involved.…”

Comparison of Motion Smoothing Strategies for Video Stabilization using Parametric Models

“…It has been proposed to stabilize the video gray scale distribution and to compensate flickering effects [11,50]. Several works have used parametric models between two successive frames in the video for the purpose of video indexing, camera motion characterization, or ego-motion classification, such as [5] or [26]. The scale-space analysis that we propose for video stabilization can also be used for other related tasks, such as activity classification and detection [24], [54], [47].…”

“…This process is first carried out on the left columns, using a temporal array, L(j) (lines 7-13), and then on the right, using another temporal array, R(j) (lines [15][16][17][18][19][20][21]. The image is processed line by line, calculating the maximum area (lines [22][23][24][25][26][27][28][29]. Using a stack, this algorithm has O(N ) complexity.…”