提出一种达到实时处理能力的自适应电子稳像系统。首先,采用基于投影相关单峰性的配准策略进行全局运动估计,为了减小投影全相关计算的复杂度,提出了快速投影算法(FPA),采用3点自适应搜索策略,结合大小搜索步长,进行由粗到细的搜索定位。其次,详细阐述了自适应均值滤波方法,通过自适应调整滑动窗口的尺寸,实现较好的平滑处理以降低抖动,并防止了过稳现象的发生。最后,提出运动矢量修正措施,以自适应地选取待补偿帧,来防止图像边缘的错误累积效应。实验结果表明,该系统可以明显减轻视频序列抖动现象,而且能实时跟随真实扫描场景。
We present a real time implementation of an adaptive electronic image stabilization system. Firstly,the global motion is estimated based on searching the single peak of gray projection correlation curves. In order to reduce the computation cost of the full correlation, a fast projection algorithm(FPA) is presented. It performs the three points' adaptive search method combining large and small search steps, to realize motion estimation from coarse-to-precise. Secondly, the moving average filter,whose length is adaptive to the dithering frequency,is used to smooth out the unwanted camera motion. This adaptive motion smoothing scheme can prevent over-stabilization and preserve the camera scanning. Lastly,the motion modification is proposed to reduce the boundary error accumulation. The experimental results show that this system can effectively alleviate the unwanted camera translation jitter and follow the intentional camera scan.