Design And Implementation Of Embedded Image Feature Extraction System Based On Zynq

Image feature extraction is the prerequisite and basis of high-level application of digital image processing technology.Fast and accurate extraction of image features is the key of image segmentation,image matching and machine vision.The existing embedded image processing system mainly uses the software to realize the algorithm of image feature extraction.Since the feature extraction algorithm needs to operate every pixel of the image,the computational complexity is huge,causing the performance and the processing speed of the whole system unsatisfactory.And if using the decentralized multi-processor architecture,it will lead to the difficulty of system development,system stability and other issues.In this paper,a Zynq-based embedded image feature extraction system design scheme is proposed,which uses hardware and software co-design method to achieve the feature extraction IP core design,hardware platform construction,embedded Linux system transplantation and application design.Zynq is consisted of ARM+FPGA.The huge image feature extraction algorithm can be unloaded into FPGA and taking the advantage of its parallel operation to improve the processing speed and performance of the system.At the same time,the flexibility of the system increases with the combination of the ARM processor.The main contents of this paper include the following four aspects:(1)The common used corner detection and edge detection algorithms are discussed in details,and then the suitable corner detection and edge detection algorithms for this paper are choosed considering the checking accuracy and the computing burdens.(2)The key technology and development flow of Zynq platform are summarized,and then the high-level comprehensive tool Vivado HLS is adopted to design and realize FAST coner detection and Sobel edge detection IP core;(3)Based on the IP core,using the custom peripherals to establish the embedded image feature extraction hardware platform which contains image acquisition,image feature extraction with hardware acceleration and HDMI displays.(4)The embedded Linux operating system is transplanted on the customized hardware platform and the software design of application layers is realized.The main works of software design include three parts,which are image acquisition,image feature extraction and display.Finally,the hardware and software integrated test of the designed system is carried out on the ZedBoard development board.The result meets the expected and it show good performace on corner extraction and edge detection.Besides,the designed system has high real-time performance.In the case of transmission delay,the FAST corner detection module takes 13.06ms for the image with size 640×480,and the processing of the Sobel edge detection module takes 11.68ms.Compared to the corresponding image feature extraction algorithm implemented in ARM,the speed of image feature extraction hardware acceleration module is three times faster than software,which achieves the purpose of hardware acceleration.The designed system can provide a relatively ideal platform for high-level application of image processing,so it has certain engineering practical value.