针对传统光流运动图像分析方法运动估计精度低且易受各种干扰等缺点,提出一种基于梯度阈值的五置信点约束的改进光流算法.该方法采用五置信点加权法处理像素时间和空间梯度值,减小了单个像素受噪声的影响.对基本光流约束方程进行了梯度阈值处理,剔除了光流场中不利于运动估计的干扰数据,提高了运动图像的运动估计精度.实验分别采用传统算法和改进算法对全局和局部运动场景的视频图像序列进行光流计算,结果表明,该改进光流算法运动估计精度高、抗干扰性强,能准确地反映场景全局运动及场景中的局部运动,比传统的光流算法具有更好的鲁棒性和收敛性,且运算时间缩短了1/3.该光流算法在行车环境传感、交通流分析中具有潜在的应用价值.%For the problems of low motion estimation accuracy and easy to be affected by various interferences such as object surface structure change, background light disturbance etc, when solving the basic optical flow constraint equation with the traditional algorithm, an improved optical flow algorithm with five-confidence-point restraint based on gradient threshold has been put forward in this paper. The five-confidence-point weighted method was applied to calculate the time and space gradient value of each image pixel, reducing noise effect on it, and gradient and gradient threshold was adopted to process the basic optical flow constraint equation, so that the interference data which will disturb the motion estimation in the optical flow field were eliminated, and the motion estimation accuracy of motion image was improved. The experiments have been carried out by the traditional algorithm and the improved one respectively for global and local movements. The contrastive analysis shows that this improved optical flow algorithm, which can figure both scene global motion and local motion in the scene, has the advantages of not only a high accuracy of motion estimation and a strong anti-inter-ference, but also a better robustness and convergence compared with the traditional one, as well as one-third of computaion time shortened. It can be expected that this algorithm will be of potential application in driving environment sensing and traffic flow analysis.
展开▼
