Abstract: In order to make the data responding to the edges of moving objects expand to the area，whose gray is flat，it needs to keep iterative numbers enough large on gradient-based optical flow algorithm.The traditional approach needs a large amount of calculation，but result accuracy is low.The gradient number accuracy and reliability will directly determine the result of optical flow algorithm.It is possible to manage gradient number effectively，modify them and improve numerical reliability of image gradient data，by using Hessian matrix distinguishing，Gaussian filtering standard deviation amending，mean model amending and multi-image comparing.These approaches can prevent gradient data of moving objects edge diffusing around，and guide them indirectly.Then，they improve precision and convergence speed for optical flow calculation.

摘要： 基于梯度方法的光流场算法中，迭代次数需要足够大才能使运动物体边界数据扩散到灰度平坦的区域，计算量大，精度不高。在光流迭代方程中，梯度数据的精确度与可信度对光流计算的结果有直接的影响。对这一问题，应用Hessian矩阵判别、高斯滤波标准差修正、均值模板修正和多帧图像对比等四种方法可以有效地处理图像梯度数据，并不断修正，以提高梯度数据的可信度。这些方法通过防止运动物体边界梯度数据向邻域内盲目扩散，增加扩散的方向性，从而提高光流场计算精度和收敛速度。