Abstract: In the high-speed image processing system, in order to solve real-time performance of the MAD algorithm, an optimized MAD algorithm based on TMS320C6416（C6416） is presented. The design uses the technologies like appropriate data storage in SRAM, efficient program flow design and writes ASM program for the loop kernel to reach parallel processing ability of the C6416. It solves the difficulty of time consumption of the MAD. The optimized algorithm can be suitable for the situation that the random changing of template size and search window size. As to the MAD process of 64×64 template and 128×128 search window, it can be proved that the C6416 spares 9 ms only.

Key words: real-time image processing, optimization of program based on DSP, C6416, MAD

摘要： 针对高速图像处理系统中相关跟踪核心模块最小绝对差累加和算法（MAD）的实时性问题，利用TMS320C6416（C6416）处理器的内部并行结构，提出了基于该处理器的MAD算法设计方案。通过合理的内存数据存储安排、高效的程序流程设计以及人工优化循环核等技术措施，充分发挥了C6416的并行处理能力，解决了MAD模块耗时过长的难题。实验结果证明，算法能适应模板与搜索区域大小可变的情况，对于64×64的模板与128×128的搜索区域计算耗时仅9 ms，大大缩短了处理时间。

关键词: 实时图像处理, DSP程序优化, C6416处理器, 最小绝对差累加和（MAD）