Design And Implementation Of Intelligent Target Tracking System Based On FPGA

The intelligent target tracking system automatically tracks the target in real time after a target is selected by the user.Generally,the PC-based target tracking system has rich development resources,low difficulty,and high performance.However,the space size of the PC is too large,the power consumption is extremely high,the customization is difficult,so the small size,low power consumption and customizable requirements of the special design requirements cannot be meet.After comprehensively considering the advantages and disadvantages of various embedded platforms,this paper selects FPGA+DSP architecture as the hardware carrier of intelligent target tracking system.The two chips cooperate with each other to give full play to their respective advantages,in the premise of low power consumption and small volume.Achieve high performance intelligent target tracking.This paper mainly includes the pre-plan customization,resource evaluation,device selection,medium-term schematic design,PCB design,production,and later FPGA chip-side program,acceleration algorithm design,complete module debugging and system joint debugging,combined with video compression / The decompression module finally realized the entire intelligent target tracking system.The system uses Xilinx's Artix-XC7A200 T model FPGA and TI's TMS320C6657 model DSP.As the main control unit and arithmetic accelerator of the system,FPGA is responsible for the underlying implementation of intelligent tracking algorithm based on machine learning,as well as system clock resource allocation,power-on reset control,timing control between modules,and image and control data transmission.The FPGA and DSP are connected by SRIO high-speed serial data channel for image/control data and tracking data interaction.The DSP chip is the main implementation chip of the tracking program,and realizes the core algorithm and algorithm flow control of the intelligent target tracking.The acceleration scheme of the intelligent tracking algorithm includes a two-dimensional FFT algorithm acceleration module in a KCF-based machine learning tracking algorithm,and an acceleration process of a convolutional layer and a pooling layer in a convolutional neural network algorithm.In the KCF-based machine learning algorithm,the FPGA uses the fixed-point two-dimensional FFT,which greatly shortens the system running time,obtains good tracking results,and accelerates the intelligent tracking algorithm.Then,using the advantages of FPGA parallel pipeline,the convolutional neural network accelerator design is implemented in FPGA,which mainly includes convolutional layer,pooling layer and active layer design.The whole convolution processing unit is completed,and one volume is realized on the DSP.The time for product and pooling operations has been reduced by 75%.