基于预选算法的全局平移运动参数估计

Global translational motion parameters estimation based on pre-selection algorithm

为了快速准确估计出视频序列中存在的全局平移抖动,分析影响全局运动估计精度的各种常见因素,归纳其特点,同时结合相关基础理论,以C-1BT变换为基础,提出一种基于预选算法的快速全局平移运动估计算法。首先,在估计局部运动矢量前,采用预选算法保留能很好代表全局运动性质的部分宏块。其次,根据预选块的特点改变传统自适应十字搜索法(Adaptive Rood Pattern Search,ARPS)的臂长预测部分,减少搜索陷入局部最优的风险。最后,采用平移运动模型得到最终参数。实验仿真结果证明提出方法的快速性和鲁棒性。

In order to estimate the camera＇s small mismatching factors were analyzed, such as self-moving amplitude translational dither motion, various objects, unstable edge details and smooth areas. After summarizing the characteristics of various factors and the existing related basic theories, and the same time on the basis of the improved C-1BT transformation, the fast global translational motion parameters estimation based on the pre-selection process algorithm were proposed. Firstly, before the local motion vector estimation, the pre-selection process was used to keep some special sub-blocks, because they can better represent the global motion nature. Secondly, according to the characteristic of the pre-selection algorithm, the initial arm-length predictive part of the traditional adaptive rood pattern search method was improved to reduce the risk of falling into local optimum. At last, the optimum global motion parameters were obtained, with the translational motion model and the statistical principles. MATLAB simulation results show that the proposed method fast and robustness.