SURF Feature Detection And Matching On ZYNQ SOC

The widespread use of embedded devices is now ubiquitous in our everyday lives,with mobile handsets,ATMs and more.The rapid development of image processing technology,there are some relevant research results in medical image processing,entertainment multimedia technology,industrial testing and military testing tracking.However,the embedded devices are limited by their power consumption and volume,and there are few of computing resources for embedded devices.But,the majority of image processing algorithms are computationally intensive and have high computational complexity.So,it is difficult to meet real-time requirements in embedded image processing devices.Therefore,it's necessary to solve real-time issues in the embedded image processing system with hardware to accelerate the image processing algorithms.In this thesis,acceleration for SURF feature point extraction algorithm with the FPGA was studied.A hardware acceleration circuit with parallel and pipelined structure was designed according to SURF feature point extraction algorithm.By optimizing the data width of the circuit and the hessian matrix calculation process,the resource consumption is reduced and the calculation accuracy is improved.The maximum operating clock of the SURF circuit reaches 100 MHz,with a processing speed of 270 fps for pictures with 640 * 480 resolution;By comparing the results of open-SURF software,the relative errors of hessian matrix response between fixed-point and floating-point are basically less than 1%.Image feature point detection system was implemented i n the Xilinx's Zed Board development board,based on SURF circuit.The system is composed of three parts: the hardware system based on SURF circuit,the embedded Linux operating system and the application with graphical interface using QT and open-CV.And the system has three functions: user-specified image feature point extraction,user-specified two images matching and extracting feature points of video stream from USB camera.The system can detect the feature points at a high speed with software and hardware co-work.In the future,other image processing applications can be developed based on the system.