Research On Embedded Real-Time Image Matching Algorithm Based On ZYNQ

With the rapid popularization of intelligent unmanned technology,during the epidemic prevention and control of new coronavirus(COVID-19),the use of this technology can effectively ensure the normal operation of society rework and reproduction,reduce the risk of cross infection contact between epidemic prevention control and key positions in industrial production.Compared with the high-performance computing platform,the existing feature matching algorithm has insufficient computing power and poor real-time performance in the embedded platform.The original algorithm is improved according to the characteristics of the embedded platform,and the processing time is accelerated under the premise of ensuring stability.The real-time target matching system of embedded unmanned vehicle based on ZYNQ is studied and designed.The main work of this paper is as follows :1.By studying the principles of several commonly used feature matching algorithms,experiments are carried out under different scenarios according to the characteristics of embedded systems.In order to improve the accuracy,retain the feature information,and solve the problem of insufficient operation ability of traditional embedded platforms in the face of complex algorithms,AKAZE algorithm is finally selected for feature point extraction,and binary BRISK descriptor is used to reduce the amount of tasks in actual calculation.2.Due to the characteristics of motion smoothing constraints,GMS algorithm is introduced to convert the dense number of matching points into high-quality matching points to complete the conversion of quantity and quality.On the basis of ensuring the robustness of the algorithm,the complexity of the algorithm is further reduced and the real-time performance of the algorithm is improved.3.Real-time target matching and tracking platform system of embedded mobile car based on ARM and FPGA.The hardware design is completed by the main controller module,the image acquisition module with two degrees of freedom steering gear,the motion control module of the car and the ultrasonic obstacle avoidance module.4.The V4L2 protocol is used to complete the real-time acquisition of the screen.The Linux image system is configured on ZYNQ-7010 device,the Opencv and QT libraries are crosscompiled to complete the transplantation and installation.The human-computer interaction interface and host computer interface of embedded equipment terminal are designed by QT,the data transmission between embedded equipment and host computer is realized by Ethernet communication protocol technology.In the final experimental test stage,the system is tested and verified by three different scenarios.The experimental results show that the proposed AKAZE-BG algorithm can achieve the desired effect in performance and speed.Compared with the original AKAZE algorithm,the speed is increased by 2.5 times,the accuracy is still better than the original algorithm.