Quantitative motion analysis from echocardiography is an important yet challenging problem. We develop a motion estimation algorithm for echocardiographic image sequences based on diffeomorphic image registration in which the velocity field is spatiotemporally smooth. The novelty of this work is that instead of optimizing a functional of velocity field which consists of similarity metrics between a reference image to each of the following images (\textitfirst-to-follow), we optimize a functional which is a sum of similarity metrics of each two consecutive images (\textitframe-to-frame). This method can reduce the bias effect of using a single image as reference. It also improves registration accuracy since consecutive frames usually have higher dependency than frames far away. We validate our method by using both simulated images with known ground truth and \textitin vivo pig heart images with sonomicrometry. Tests indicate that our frame-to-frame motion estimation method is more accurate than first-to-follow method.