Embedded vision system is small, low power consumption, easy to install and easy to maintain, which is used increasingly wide in industrial machine vision and intelligent robot area, but it still has a certain performance gap compared with vision system based on PC platform. We can improve the performance of embedded vision system from both software and hardware aspects: on software aspect, image processing algorithms can be improved to increase the robustness and computational efficiency of the vision system; on hardware aspect, hardware structure can be improved, we can design parallelized hardware architecture for image processing algorithms. In this paper, supported by National 863 Projects, hardware structure and software realization are studied and researched on an embedded vision system based on FPGA and DSP. The novel work and contributions of this thesis includes: Firstly, a parallelized image processing structure on FPGA is proposed for the embedded image processing card designed by our research team. The portability of image processing modules is improved by introducing external memory access interface based on Wishbone bus into the structure. Secondly, according to the characteristics of neighborhood algorithms, a hardware structure is designed and a lot of neighborhood image processing algorithms are realized on FPGA, including gray scale transformation, median filtering, Gaussian smoothing, corrosion, expansion, Sobel operator and so on. Canny edge detection algorithm is improved and realized with a parallelized structure on FPGA, the result of edge detection is perfect. Thirdly, with FPGA and DSP cooperating with each other, the butterfly color tag and MR code recognition methods are realized on the embedded image processing card, which provide visual basis for the localization and navigation of embedded mobile robot. Fourthly, a pose measurement device based on dot lasers and embedded image processing card is designed for the spraying robot to measure its end position and pose. A real-time U-shape bed localization method is realized on the embedded image processing card to enable the embedded intelligent wheelchair docking into a U-shape bed automatically through visual servo.
修改评论