视频稳像评价方法研究

为有效地完成视频稳像评价任务,根据摄像头状态及场景中有无运动目标,将视频分为四类,并对每一类视频的稳像评价难点进行了分析。针对摄像头运动的条件下,峰值信噪比(peak signal to noise ratio,PSNR)无法正确评价视频稳像效果的问题,提出了基于全局运动矢量角变化(variation of intersection angles,VIA)的方法来衡量两段视频的稳定程度。由于场景中运动目标对PSNR及VIA的计算均有影响,针对固定背景存在运动目标的情况,提出了基于背景重叠区域的PSNR方法;而在动态背景条件下,通过筛选运动目标上的运动矢量,有效去除前景运动目标对于全局运动矢量计算的干扰。实验证明,提出的评价体系,能够在不同类别的视频中去除运动目标带来的干扰,正确评价各类视频的稳像效果。

Research on evaluation method of video stabilization

In order to accomplish the task of evaluating video stabilization, this paper divided videos into four classes according to the state of the cameras and moving targets in monitoring areas. It also analyzed the difficulties of evaluating video stabilization in each condition. As PSNR(peak signal to noise ratio) works bad with moving cameras, this paper proposed a novel evaluation method based on the VIA(variation of intersection angles) between global motion vectors. In order to decrease the influences to PSNR caused by moving objects in monitoring areas, the paper computed the PSNR of the overlapping backgrounds instead of the whole image achieved by fixed camera. As moving objects also affect VIA, the paper deleted the motion vectors on moving objects and removed the related influence. As the experiments show, the evaluating architecture proposed in this paper can remove the disturbance of moving object in each class of video and evaluate video stabilization correctly.