摘要
为了解决传统HDL语言编写现场可编程门阵列(FPGA)部署神经网络算法时存在的开发难度高、周期长、可移植性差的问题,设计了一种基于PYNQ框架的人体目标跟踪系统。该系统将加速的SSD算法部署在低功耗ARM+FPGA的异构计算平台Mizar Z7开发板上,对基于卷积神经网络的SSD算法进行软硬件协同开发。PL端设计卷积层加速器,移植PYNQ框架后通过PS端Jupyter Notebook调用综合设计完成ARM与FPGA间高速的信息交互,实现对摄像头采集的图像进行人体目标检测与运动轨迹显示的功能。经过实际测试,该系统可以实现实时识别人体目标、采集人体运动轨迹的功能,可适用于人体目标跟踪相关领域。
In order to solve the problems of high development difficulty,long cycle and poor portability in programming neural network algorithm with traditional HDL language,a human body target tracking system based on PYNQ framework was designed.In this system,the accelerated SSD algorithm is deployed on Mizar Z7 development board,a heterogeneous computing platform based on ARM+FPGA with low power consumption,and the SSD algorithm based on convolutional neural network is co-developed with software and hardware.The convolution layer accelerator is designed on the PL side,and after the transplantation of PYNQ framework,the high-speed information interaction between ARM and FPGA is completed through the comprehensive design of Jupyter Notebook call on the PS side,so as to realize the function of human target detection and motion track display on the images collected by the camera.After practical testing,the system can realize the real-time recognition of human target and acquisition of human movement trajectory,which can be applied to the field of human target tracking.
作者
卫建华
刘润利
许佳豪
尚晓峰
Wei Jianhua;Liu Runli;Xu Jiahao;Shang Xiaofeng(School of Electronic Information,Xi′an Polytechnic University,Xi′an 710048,China)
出处
《国外电子测量技术》
北大核心
2021年第12期89-95,共7页
Foreign Electronic Measurement Technology