| With the explosion of image processing,computer vision and other related concepts in recent years,image stitching has also become one of the popular research directions under the influence of image retrieval field and computer vision field.SIFT algorithm,as a foreground detection algorithm based on feature recognition,is often used in image stitching,image retrieval,feature extraction,image recognition and other fields by virtue of its rotation,translation,scaling invariance and resistance to light brightness changes,etc.Image recognition and other fields.However,it has the shortcomings of large computation and long computation time,which limit its usage scenarios.In this thesis,we design a new and improved method to shorten the computation time by moving it to hardware and using FPGA as the core of hardware computing.The main work of this thesis is listed as follows.(1)From the research results and system requirements,the system implementation scheme is divided into two parts: hardware and software,and the functions to be achieved by each part of the module are divided,after which the necessity of adopting FPGA as the hardware platform is demonstrated.(2)Build a two-way image acquisition system and use it as a simulation of the onboard display multiplex synchronization system.The adapter board is designed to fix two monocular cameras,while adjusting the relative positions of the two monocular cameras for the purpose of expanding the viewing angle.Software level design camera driver,DDR3 driver,HDMI display driver,design image ping pong cache structure,dual camera read/write arbitration.Write the driver according to SCCB protocol and AXI4 communication protocol.Finally,the synchronous output of two 640×480@60 Hz images is completed and displayed on the left and right halves of the HDMI display respectively.(3)Analyze the SIFT algorithm,propose an algorithm optimization scheme,perform MATLAB simulation on the optimized algorithm to verify the reasonableness of the algorithm,and determine the parameter selection,filtering method,matching method,etc.in the final algorithm after multiple comparisons.Based on the MATLAB simulation,the hardware implementation of each module of the algorithm is designed considering the limitation of hardware self-contained resources and the characteristics of FPGA itself with high concurrency and shaping calculation.This paper highlights several processing methods and techniques commonly used in hardware logic design,such as Gaussian kernel separation,image window construction,data caching,cyclic shifting,etc.(4)Based on the dual-channel image acquisition system,the top-level connection of the whole system is designed with emphasis on reducing the resource consumption.Including image cropping module,RGB image to grayscale module,etc.to pre-process the image.Write the top-level algorithm verification module to test the system.(5)Analyze the resource consumption of each module and the deficiencies in the experiment for secondary development and optimization.Finally,the two-way synchronous acquisition system built in the system test runs smoothly without lag,and the use of two-way camera expands the view range to achieve the effect of driving assistance;the simulation results of the improved algorithm show that for the same two images,the average time consumed by the original algorithm in MATLAB to match feature points is 5.11 s,and the average time consumed by the optimized algorithm in MATLAB to match feature points to output the stitched image after the whole process is 1.08 s,and 4.171 ms for matching feature points on FPGA,the time efficiency of the improved algorithm is improved by 373%. |
PDF Full Text Request