基于多核DSP激光成像雷达数据处理系统

采用多核DSP设计了一个用于地面目标检测的激光雷达实时图像处理系统。在详细分析算法各模块资源消耗量的基础上,完成了硬件电路设计,实现了以主辅拓扑结构为框架的软件并行处理系统开发。在系统实现时,先将图像进行分区,并合理地将分区后的图像分配到四个DSP核中进行处理。最后,将并行系统进一步扩展到双核和六核,并与单核系统进行性能比较。对算法运算时间的测试结果表明,系统处理一帧图像仅需50 ms达到了实时性要求。结果表明,对于固定负载的处理系统,单纯地通过增加并行的核数来提高加速比的幅度是有限的。当增加并行的核数已不能明显地提高计算效率时,在系统设计中应着重减少每个核串行运算的负载量。

LiDAR data processing system based on multi-core DSP

A real-time LiDAR image processing system for ground target detection based on multi-core Digital Signal Processors (DSPs) is proposed.On the basis of detailed analysis of the source consumption for each algorithm module,the hardware circuit for the algorithm is then designed,and a parallel processing system is implemented using the master-slaver architecture.Each image is first divided into several partitions,and each partitioned image is then assigned to one of the four DSP cores for further processing.Moreover,the parallel system is then extended to a dual-core system and also a six-core system.Their performances are further compared with that of the single-core system.Experimental results show that it only takes 50 ms to process a frame image by using the designed system,so it can meet the real-time requirement.For processing system with fixed computational load,the improvement in computational efficiency will be limited if only the strategy of increasing the number of cores is adopted in the system.Once the computational efficiency cannot be further improved by adding more cores,research emphasis should be shifted towards the direction of how to reduce the computational load in each core.