摘要
基于MES50HP开发板及PC主机实现了多源视频采集、拼接及AI协同处理的加速工作。所提出的系统可支持同时采集HDMI、网口、摄像头和光纤4路视频数据,视频采集拼接后,将缩放后存入双倍速率同步动态随机存储器(DDR)中,输出部分分为2路,其中1路数据用于HDMI回环输出,另外1路通过PCIE传至PC主机用于结果显示,并在PC端读取PCIE传回的数据并显示目标检测结果,其中视频采集和目标识别加速部分主要由2块MES50HP开发板构成。本系统在FPGA上实现了卷积加速器,卷积加速器计算完神经网络的一层后再通过PCIE传至主机进行结果显示,卷积计算与图像采集进行深度融合,具有实时、低成本的特点,可广泛应用于边缘计算等领域。实验结果表明:针对红绿灯数据集,该方法在低成本PC主机上实现的最大平均精度均值(mAP)为0.746,最高帧率达45帧。
This study realized the acceleration of multi-channel source video collection,splicing,and AI collaborative processing based on the MES50HP development board and PC host.The proposed system can support the simultaneous collection of four-channel video data from HDMI,network port,camera,and optical fiber.After the video is collected and spliced,it will be scaled and stored in double data rate synchronous dynamic random access memory.The output part is divided into two channels,one of which is used for HDMI loop-back output,and the other is transmitted to the PC host through peripheral component interconnect express(PCIE)for result display,and the data returned by PCIE is read on the PC side and the target detection results are displayed.The video acquisition and target recognition acceleration parts are mainly composed of two MES50HP development boards.This system implements a convolution accelerator on FPGA,the convolution accelerator calculates one layer of the neural network and then transmits it to the host through PCIE for the results display.The convolution calculation is deeply fused with image acquisition,which is real-time and low-cost and can be widely used in edge computing and other fields.The results showed that for the traffic light data set,this method achieved a maximum mean average precision value of 0.746 and a maximum frame rate of 45 on a PC host.
作者
吴铭
黄国宁
汪保祥
宋可平
鄢秋荣
吴武飞
WU Ming;HUANG Guoning;WANG Baoxiang;SONG Keping;YAN Qiurong;WU Wufei(School of Mathematics and Computer Sciences,Nanchang University,Nanchang 330031,China;School of Information Engineering,Nanchang University,Nanchang 330031,China)
出处
《南昌大学学报（工科版）》
CAS
2024年第3期386-394,共9页
Journal of Nanchang University(Engineering & Technology)
基金
国家自然科学基金资助项目(62002147)
中国博士后科学基金资助项目(2020TQ013)
江西省自然科学基金面上资助项目(20232BAB202009)。
关键词
现场可编程门阵列
外设组件互联快速总线
AI加速
视频采集
边缘计算
嵌入式人工智能
field-programmable gate array
peripheral component interconnect express
AI acceleration
video capture
edge computing
embedded artificial intelligence
