摘要
针对传统DsP＋FPGA设计方案结构复杂、功耗高、体积大、抗干扰性差等问题，应用可编程片上（SOPC）技术建立了多目标实时跟踪系统并提出了基于NIOSII核及其数据链路特征的软件优化方法，对系统采用的迭代投影跟踪算法和通信制导控制算法进行了优化。研究了算法结构优化，链路配置优化，代码优化，自定义指令和C2H技术，并将这些技术应用于多目标跟踪SOPC系统来优化算法结构，减少代码量，去除程序相关性并进行链路层硬件转换，提升软件并行流水性能。实验结果表明，该优化方法可使软件效率提升9～18倍。在图像序列100frame／s，分辨率256pixel×256pixel的条件下，在1．3ms内完成了单帧图像的多目标跟踪处理，470us内完成了单周期导航运算处理，使系统运算满足了实时处理的要求并在节约硬件成本的同时了简化结构，提高了运行稳定性。
For the problems of complex structure, high power consumption, large size and poor antiinterference in the traditional DSP＋FPGA design, a real-time multi-target tracking system based on a System on Programmble Chip（SOPC） was established and a software optimization method based on NIOS II core and its data link was proposed. The optimization technologies for algorithm structure, data link configuration, code optimization, custom instruction and C2H were researched. Then, these technologies were applied to the multi-target tracking SOPC system to optimize algorithm structure, reduce the size of code, remove the correlation of code and implement link-layer hardware conversion to enhance the performance of software pipelining. Experimental results indicate that these optimizing technologies can improve the efficiency of software by 9-18 times. The computing time of multi-target tracking is 1.3 ms/frame and the navigation operation time is 470 us/cycle for the image sequence of 256 pixel × 256 pixel in 100 frame/s. The system can realize real-time processing with low hardware costs, simple structure and stable performance.
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2011年第3期681-689,共9页
Optics and Precision Engineering
基金
国家863高技术研究发展计划资助项目(No.2008AA8012320B)
关键词
软件优化
多目标跟踪
实时处理
可编程片上系统
software optimization
multi-target tracking
real-time processing
System On Programmable Chip（SOPC）
