由于理论上的局限性, 逆向复合模板匹配算法难以在末制导阶段的高动态环境下实现对目标的稳定跟踪。针对这一问题, 本文提出一种基于多分辨率运动先验的逆向复合跟踪算法。算法将跟踪问题分为离线训练和在线跟踪两个阶段。在离线计算阶段, 将目标运动的幅值按 “coarse-to-fine”的顺序划分为多个层次 ,并采用分层训练的方式获得多分辨率先验误差雅可比矩阵。将该先验知识应用于逆向复合跟踪算法, 能够在不增加在线计算复杂度的前提下实现对地面固定目标的实时稳定跟踪。对比实验证实, 在高动态环境下, 算法具有良好的收敛特性, 同时, 对目标图像的旋转、尺度缩放和光照变化等干扰因素也具有良好的稳定性。

During terminal guidance phase of electro-optical precision-guided weapons, basic inverse compositional template matching algorithm can not deal with robust target tracking in high-dynamic environment due to its theoretical limitations. An inverse compositional tracking algorithm using multi-resolution motion prior is proposed to deal with this problem. The novel method divides tracking problem into two phases of off-line training and on-line tracking. During training phase, multi-resolution priori error Jacobian matrices are obtained according to the principle of “coarse-to-fine”. Using this priori knowledge, the new tracker can achieve robust tracking without increasing on-line computation complexity. Comparative experiments confirm the good convergence properties of the algorithm under high-dynamic environment. Meanwhile, the algorithm also has a good stability against the disturbances of target image rotation, scale and illumination changes, etc.