Quantitative characterization of an image brightness subtraction method for motion determination.

摘要

We present a method for estimating the average frame-to-frame motion of a videotaped textured surface undergoing deformations. The method is based on subtracting successive frames from the video sequence and generates an approximation to the spatial average of the square of the surface's displacement field between the two frames. The method is computationally much faster than a full optical flow analysis, but is not always quantitatively accurate. We identify the parameters governing the errors, identify the regime in which the method is quantitatively accurate, and demonstrate that the accuracy can be assessed using only information internal to the algorithm. Finally, we apply both our method and a correspondence-based optical flow method to video sequences of spontaneously beating tissue cultures of embryonic chick cardiomyocytes. We show that the extent of the agreement between the two sets of results matches the predicted values.