模型基编码中实时三维运动估计的新方法

New real-time 3D motion estimation in model-based coding

将自适应迭代松弛搜索算法RSA引入到MPEG4模型基编码的三维运动参数估计中 ,解决了估计算法的稳健性问题 .经过简化计算的复杂度 ,得出一种快速算法 ,大大提高了估计速度 .同时设计有效的参数初值设置和调整方法 ,改善了估计的准确性 ,使之更适用于长序列实时运动估计 .通过与预测最小均算法PLS和扩展的卡尔曼滤波算法EKF进行模拟数据和真实序列实验比较 。

The problem of robustness in the estimation algorithm is solved by applying relax iterative search algorithm (RSA) to motion estimation of MPEG4 model-based coding. And then, a faster computation is achieved by reducing algorithm complexity. The efficient method of setting and adjusting parameters highly improves the estimation precision, making the RSA more suitable for long-term real-time estimation. Compared with the predictive least squares algorithm (PLS) and extended Kalman filter (EKF) in the experiments using synthetic data and real image sequences respectively, the results prove that the proposed scheme RSA is robust and accurate.