【摘要】 针对传统的基于梯度的光流方法一般不能检测出帧间位移小于1个像素的运动目标,提出一种改进的基于梯度的光流计算方法.该方法使用了一种新的3DSobel算子计算像素的时空梯度,并增加置信判断算子来确定每个像素对应的光流场,改进的梯度计算过程需要5帧图像.实验结果表明在静止背景的图像序列中,该方法比传统方法能更好地检测出以不同速度运动的目标.更多还原

【Abstract】 Considering the traditional optical flow algorithm couldn′t detect the moving objects in case the displacement between two continuous frames were less than one pixel, we developed an improved gradient-based optical flow algorithm that separates continuous frames into two groups. The optical flow fields of the middle frame were computed respectively. The process of gradient computing used 3D-Sobel operator. A parameter was defined to determine the final optical flow field of the middle frame. The results show that the algorithm can accurately and validly detect different objects with different velocities in static background.更多还原