A motion estimation algorithm based on gradient methods for low signal-to-noise (SNR) scenarios is presented by using statistical performance of the estimator.Firstly

the cost function of mean square error (MSE) is developed based on Cramer-Rao low bound by considering the influence of the noises on motion estimation.Secondly

the motion estimation MSE is minimized to find the gradient optimal filters.In combination with multi-scale pyramid approach

the estimator accuracy of such an algorithm can be further improved.Experimental simulations show that the estimator bias is less than 0.008 pixels for large motion estimation of low SNR scenarios.This represents a significant decrease in estimator accuracy compared to existing methods for motion estimation of low SNR situations.