基于多参量模型的全局运动估计算法

提出了一种基于特征点匹配的视频序列多参量运动估计算法,即将初始特征点域,进行梯度方向上的极坐标变换,并依据正交特征,作方向轴上的一维投影,构建出新的特征曲线,从而将孤立的特征点匹配转化成特征曲线相关.依据曲线的相关性,完成特征点对的匹配,最后使用最小二乘解算平移、旋转、缩放等变换参量.实验结果表明:算法的特征点误匹配率<5%,缩放误差<0.1%,平移误差<0.1像素;各参量估计范围:缩放因子s:0.7130°,平移参量|d|>35.算法在保证了高精度的同时,具有更为宽松的使用条件及适用性.

Multi-parameter Globe Motion Estimation Algorithm

To estimation the multi-parameter including scaling, rotation and translation, a robust globe motion estimation algorithm was proposed. In this algorithm, feature point matching is carried out through with feature curve matching, which mainly includes three steps: constructing feature block, creating feature curve and calculating feature curve correlation. Then, a linear system is constructed and solved by repeated least-squares to generate the global motion parameters with these matched feature point pairs. Experiments show that, the matching error of corresponding point pair is less than 5%, and the multi-parameter error is less than 0.1 pixel, 0.1degree, and 0.1% scales. The processing capability: scaling varying from 0.71 to 1.41, rotation factor R>30 degree, translation factor |d|>35. The algorithm is robust and can be widely applied to global motion estimation field.