基于PYNQ的运动目标检测系统

Moving object detection system based on PYNQ

运动目标检测技术作为计算机视觉领域的研究热点之一,在智能安防、重点场所监控等领域应用广泛。为了在PYNQ-Z2开发平台上实现运动目标检测系统,基于软硬件协同设计的思想进行了系统功能划分。在可编程逻辑PL部分,使用Xilinx Vitis HLS工具进行图像处理模块硬件加速IP核的开发和整合。在处理系统PS端,通过Jupyter Notebook平台使用Python语言进行视频数据的读取以及硬件的调用。实验结果表明,在PYNQ-Z2平台上实现了运动目标检测系统,可同时对多个运动目标进行检测,在输入视频分辨率为800×600时的处理速度为14.3帧/秒,基本能满足实时性要求。

As one of the research hotspots in the field of computer vision,moving object detection technology is widely used in intelligent security,special place monitoring,and so on.In order to implement the moving object detection system on PYNQ-Z2,the system function is divided based on the idea of hardware⁃software codesign.In the programmable logic named PL,Xilinx Vitis HLS tool is used to develop and integrate the IP core of each image processing module.On the processing system named PS,the video data is read and the hardware is called by using Python through Jupyter Notebook.The experimental results show that the moving target detection system can be implemented on the platform of PYNQ-Z2,which can detect multiple moving targets at the same time.When the input video resolution is 800×600,the processing speed is 14.3 frame/s,which can basically meet the real⁃time requirement.