基于光流法的多分辨率电子稳像算法

Multi-resolution Electronic Image Stabilization Approach Using Optical Flow Algorithm

摘要:

针对动载体摄像系统中视频序列受载体姿态运动及抖动的干扰而出现的不稳定现象, 提出一种基于光流算法的多分辨率电子稳像算法。首先, 通过划定有效的运动估算区域取代对整帧图像的计算以降低计算量; 然后, 利用基于光流算法的多分辨率分层运动估计快速并精确地计算出包含平移、旋转以及缩放运动的相邻帧间仿射变换参数; 最后, 采用固定帧补偿算法, 利用求得的仿射变换参数, 对图像进行运动补偿, 消除或减轻图像序列帧间的随机抖动, 达到稳像的目的。实验结果表明, 针对包含运动目标的动态场景, 该算法可以精确地检测出视频序列帧间平移、旋转以及缩放等复杂的抖动, 水平和垂直方向的稳像精确度小于1 pixel, 保证视频序列的稳定输出, 可应用于目标跟踪系统中。

Abstract:

To overcome the unstable video sequence in moving vehicle camera system, a multi-resolution electronic image stabilization approach based on optical flow technique was proposed. First, the calculation amount was reduced by drawing effective motion estimation area instead of making calculations on the whole frame. Then, multi-resolution hierarchy motion estimation based on optical flow technique was used to calculate the affine motion parameters of inter-frames fast and precisely. Finally, the obtained affine transformation parameters were used to compensate the images with fixed frame compensation algorithm. The random jitter in inter-frames of image sequence was removed, and image stabilization was achieved. The experimental results demonstrate that the proposed approach can estimate the complex jitters including translation, rotation and scaling precisely. The stabilization accuracy in horizontal or vertical direction is lower than 1 pixel. The proposed approach can output stable video sequence, and can be applied in target tracking system.