Design Of Front-end Machine Vision System Based On Hisilicon Hi3559

With the promotion of the strategic layout of national industry 4.0,automation and intelligence have become the main direction of China’s industrial upgrading,and machine vision system plays an important role in the process of industrial upgrading.With the continuous improvement of the resolution and frame rate of CMOS image sensor,a large number of real-time image data transmission problems are brought.The existing solution is short-distance transmission,that is,to improve the bandwidth by shortening the transmission distance.Typical image transmission interfaces such as camera link can reach the transmission rate of 6.8 Gbps within 7-10 m,USB3.0 can reach the transmission rate of 5 Gbps within 5 m,and CoaXpress can reach the transmission rate of 6.25 Gbps within 90 m.The vast majority of industrial measurement sites do not have the conditions to install image processing workstations,so this short-distance real-time image data transmission method has brought great restrictions to the image detection system in industrial sites.In order to break through this technical bottleneck,this paper proposes a front-end machine vision system,which can process a large number of image data in real time,and transmit the processing results or semi-finished products to the outside,so as to achieve great data transmission reduction.The main contents of this paper are as follows:(1)Analyze the functional requirements of the system,and select the hardware on this basis.The media memory management mechanism and image acquisition function of Hisilicon hi3559 platform are studied,which lays a theoretical foundation for the implementation of subsequent related functions.Finally,the development environment of Hisilicon platform is built.(2)Based on Hisilicon Hi3559 platform and Sony imx174 image sensor,by studying the adjustment mechanism of the exposure time and gain of the image sensor,the automatic exposure function of the system is designed and realized.The gamma correction function of Hisilicon platform is realized by looking up the table.On the basis of automatic exposure,the ROI frame rate adaptive function of the system is designed and implemented,and tested.Finally,the H.264 coding function based on Hisilicon platform is realized,and the network transmission of real-time code stream is realized through RTSP protocol.(3)The algorithm download function based on NFS is realized based on Hisilicon platform.The image processing algorithm of the upper computer can be downloaded to the front end for operation.The output redirection function of the system image processing results is designed and implemented.On this basis,the software design of the video input module is completed.Based on the image acquisition process of the VI module,the software framework of front-end image processing is designed by using multithreading technology.On the basis of this framework,developers can flexibly deploy image processing tasks.Finally,the function design and implementation of spot centroid calculation are completed based on this framework.After testing,the system can calculate the spot centroid coordinates of each frame in real time at the maximum frame rate of 1018 FPS,and transmit the results to the host computer.In this paper,the concept of front-end machine vision system is proposed and verified.On the premise of retaining high resolution and frame rate,long-distance signal transmission is realized,which provides an effective way to break through the limitations of the installation conditions of industrial field machine vision system.