基于DSP的KLT特征点跟踪算法优化实现

Optimized Implementation of KLT Feature Points Tracking Algorithm Based on DSP

针对C6000系列定点DSP下KLT特征点跟踪算法的快速实现问题,通过结合DSP体系结构特点以及软件流水技术,提出了一种该类型DSP下的算法优化实现方案.基于算法流程的模块化分析,完成了该算法的定点化设计与实现工作;针对算法中计算密集型模块,提出了具体的改善软件流水,减少存储器访问,任务级并行设计等优化实现方法.在DSP软仿真平台CCS下进行实验测试,该优化实现方案下的运行效率和资源利用率均有较大的提升,且所使用的算法优化方法适用于所有C6000系列DSP.

The paper proposes an optimized implementation of the algorithm based on the platform of C6000 DSPs, in consideration of the DSP hardware resource characteristics and software pipelining technique. Based on the analysis of algorithm process modular. We completed the designation and implementation of Floating-to-Fixed-Point conversion; for Compute-Intensive modules, we present the concrete optimized method, including improving the performance of Software pipelining, reducing memory access, achieving task level parallelism. The experiment made in Code Composer Studio results show that, the efficiency and resource utilization of the optimized implementation scheme are increasing, and the optimization method in this paper also apply to all C6000 DSPs.