Target Detection And Tracking Based On FPGA

Target detection and tracking technology is not only a challenging technology in the field of machine vision,but also the most important technology branch of intelligent robots.Serving as the visual system of intelligent robots,it is responsible for the execution of autonomous positioning,environmental recognition,obstacle detection,target tracking and other bionic functions,which is a key step in the human pursuit of bionic robots.At present,the intelligent robot system has extremely high requirements for real-time and convenience,especially in the machine vision direction.Traditional machine vision systems using computers as processors have been unable to meet the requirements of robot vision systems due to the drawbacks of high data,high energy consumption,large size,and long delay.This thesis proposes a method based on FPGA platform for image processing.It utilizes the features of high real-time performance,low power consumption,parallel processing and strong scalability of FPGA to transplant machine vision algorithm hardware into FPGA,and finally achieve target detection and target tracking.In this thesis,problems of hardware selection,system function module design,system platform construction,hardware processing transplantation or acceleration of image processing algorithms were studied and solved.Finally,through the tracking experiment of visual system and car,the advantages and feasibility of using FPGA as a hardware processor to process images were verified.The main work of this thesis can be subdivided in to three parts including hardware system platform construction and modular design,image preprocessing algorithm and hardware extraction and acceleration of target extraction algorithm,and hardware implementation of target tracking algorithm.Details are as follows:1)Functional modules were designed according to system functions which mainly include VGA display,SDRAM storage and camera modules.Problems ofhardware system construction and modular design were solved and improved,including ping-pong storage design to solve frame interlacing,SCCB interface for camera configuration,and camera module read-write data design.Finally,a hardware system platform was built by assembling the OV7725 camera,SDRAM memory,VGA display and FPGAchip to form an image acquisition and reality system.2)Hardware migration of the image pre-processing algorithm at the bottom of the pixel was completed.VHDL language was applied in transplanting the algorithm to the FPGA to operate a series of color conversion which means converting the original image into RGB image,converting RGB image into YCbCr image and converting YCbCr image into grayscale image.Fast sorting median filtering algorithm was adopted to filter grayscale image and simulate Modelsim waveform.Sobeloperator edge detection method was used to realize convolution and binary of pixel point third-order matrix and operator template.After processing,the black and white image after edge detection was obtained.3)Hardware migration of the target detection and tracking algorithm for continuously moving objects was realized.The car was controlled to track the position of moving objects.The inter-frame difference algorithm is implemented and transplanted by the program,the edge contour of the moving target was extracted,the edge feature of the moving target was selected to perform the tracking algorithm based on feature matching,and the centroid position of the moving target was calculated.Finally,the car was controlled to track the real-time centroid location.The use of FPGA as a hardware processor for target detection and tracking overcomes the difficulties of real-time and portability.In today's rapid development of artificial intelligence technology,it has certain economic benefits and broad application prospects.