一种具有自适应权值系数的全局平滑光流算法  

Global Smoothness Optical Flow Method with Adaptive Weight Coefficient

作　　者：原磊[1] 李金宗[1] 李冬冬[1] 

机构地区：[1]哈尔滨工业大学航天学院,哈尔滨150001

出　　处：《数据采集与处理》2007年第2期190-195,共6页Journal of Data Acquisition and Processing

摘　　要：分析了Horn-Schunk方法在运动边界处,光流场不能很好地保持不连续性的原因,并从3个方面对Horn-Schunk迭代模型作了改进:(1)在能量方程中用可变的权值系数代替原来的常权值系数;(2)采用一种新方法求解迭代方程中的速度均值,新方法体现了邻域的亮度差别对速度扩散的影响;(3)引入补偿迭代方法去求解相关的Euler-Lagrange方程,实验证明这种迭代方法比Gauss-Seidel方法更加有效。试验结果验证了改进的光流求解模型提高了光流场的计算精度,并能更好地保持光流场在运动边界处的不连续性。A modification of Horn-Schunk approach is investigated and it leads to a better preservation of the discontinuities in optical flow field. Horn-Schunk model is improved as follows： （1）The constant smooth weight coefficient is replaced in the energy equation by a variable weight coefficient. （2） A new method is used to compute the velocity mean. The new method reflects the effect of the intensity difference on the image velocity diffusion. （3） The compensating iterative method is more efficient than the Gauss-Seidel method to solve the associated Euler-Lagrange equation. Experimental results proves the better profermace of the improved method on discontinuity preserving.

关 键 词：序列图像 光流 峰值信噪比 补偿迭代 

分 类 号：TN911.73[电子电信—通信与信息系统]