Design For Embedded Image Feature Point Extraction System On All Programmable System On Chip

This paper proposes a solution to design an embedded image feature point extraction system base on AP-SoC. The whole complete solution includes the design of the bottom hardware layer, the construction of the hardware platform system, the establishment of embedded Linux operating system, the preparation of the Linux device drivers and the implementation of the top application. The whole system is based on a Zynq-7000 AP-SoC named Zedboard which features powerful configurable logic and an embedded ARM processor. The paper adopts the ideas of hardware/software co-design. We describe the custom module that implements the Harris feature point extraction algorithm on FPGA. The module connects the processer system with ARM micro by AXI4 bus to exchange data. Finally, the module embeds a Linux distribution that allows to running Opencv programs for two enginerring applications that homologous image registration parameter extraction and target recognition.The main contributions of this work are:Firstly, this paper design a module that implements the Harris feature point extraction and builds the hardware platform system. On the basis of Harris algorithm analysis, the module can divide sub-module function reasonably with the idea of co-design and the parallelism the algorithm itself. The whole module can extract 1024 feature points mostly. By using the high-performance date port(HP)on Zedboard and CDMA IP core, the system builds a channel between the processing system and the custom module, in order to achieve an effitient hardware platform.Secondly, this paper has achieved the Linux device drivers for the custom modules. After Linux embedded operating system successfully has been established on Zedboard, we achieve the Linux device driver of the peripheral module on hardware system using the platform device mechanism, in order to provide the sofeware interface for advanced applications control the cusom mudule. The Linux drive adopts the mode of shared memory which can exchange big data between the kernel space and the user space efficiently.Finally, this paper achieves the top application by using Opencv function library, in order to verify that the system is correct. After extracting the feature points in hardware, this paper achieves the desicription of the image feature points, matching and parameter calculation by Opencv. It has proved the validity and engineering practicality of the system.This paper has implemented the image feature point extraction system. In the case of no operating system, it takes 2.9 ms to extract feature points on an image sized 512×512. Compared with the Matlab, the speedup ratio reaches 1200. With a Linux operating system, the time still can reach 5.7 ms. It means that the system has good real-time performance. In addition, the system can achieve flexibly the embedded computer vision applications on the basis of the Harris feature point extraction algorithm by using Opencv.