文中给出了一种用于实现多分辨率运动估值算法后阶段任务的改进的树结构。在一个简单的 RISC类型核控制下 ,它能够完成整个运动估值过程中除粗分辨率精度运动矢量搜索之外的所有后阶段子任务。包括运动矢量优化 (搜索 )在内的多任务是通过二叉树最底层叶节点上的多功能处理单元和可以拆分成子树的加法树来实现的。此外 ,运算单元寄存器堆的设计使能在二维方向上复用图像数据 ,完全避免了同一类数据从存储器中重复读取 ,从而实现了最小的存储器访问带宽 ,并有助于减小存储功耗 In this paper, an improved tree structure for the post-phase tasks of multi-resolution motion estimation algorithm is given. Under a simple RISC-type kernel control, it performs all post-stage subtasks except the coarse resolution precision motion vector search throughout the motion estimation process. Multi-tasking, including motion vector optimization (search), is accomplished by a multi-processing unit on the lowest leaf node of the binary tree and an addition tree that can be split into subtrees. In addition, the design of the arithmetic unit register file enables image data to be multiplexed in two dimensions, completely avoiding the same type of data from being repeatedly read from memory, resulting in minimal memory access bandwidth and reduced memory power Consumption