基于FPGA的盲人辅助视觉算法研究

发布时间：2018-05-17 06:39

本文选题：运动检测 + 背景更新策略　；参考：《西安理工大学》2017年硕士论文

【摘要】：人们常说“眼睛是心灵的窗户”。因为先天或后天原因,成千上万人的无法看到世界。盲人的日常生活受到很多限制,如行走时无法及时察觉到危险,从而给盲人出行带来严重问题。随着计算机技术和通信技术的高速发展,盲人辅助视觉设备快速发展,越来越多的科研人员致力于相关技术的研究。本文主要研究动态目标检测和静态目标边缘检测两个方面,完成以下工作:首先分析各种传统的算法原理和适用场景,采用适用于盲人行走辅助的改进算法,在动态目标检测方面采用ViBe算法结合计数更新背景和计数重新初始化背景模型机制,在静态目标边缘提取采用自适应阈值方式。然后对改进算法与传统算法进行MATLAB仿真,分别用F_Measure参数和主观观测两种方式进行比较,MATLAB结果表明,动态目标检测改进算法的召回率比原算法高10%,准确率比原算法高6%,静态目标边缘提取改进算法可以有效提取感兴趣边缘信息,排除无效干扰;接着对改进算法进行Verilog硬件电路实现,对硬件电路进行ModelSim仿真分析;最后搭建目标检测系统,从摄像头配置到VGA显示,在开发板上验证整体系统的处理效果。将硬件实现结果与MATLAB处理结果做对比,硬件处理结果相比于软件处理结果在像素点处最大相对误差为4. 3%,硬件实现达到设计要求。论文目标检测实验结果表明:动态目标检测结合ViBe算法与计数更新背景模型和计数重新初始化背景模型策略可以更适应于运动场景的目标检测;静态目标边缘检测采用自适应阈值可以减少无关信息的干扰。将目标检测通过硬件电路实现,更好满足了实时性要求,丰富了视觉假体中图像处理的研究内容。
[Abstract]:People often say, "the eyes are the windows of the soul." For congenital or acquired reasons, thousands of people can not see the world. The daily life of blind people is restricted by many limitations, such as the inability to detect danger in time when walking, which brings serious problems to the travel of blind people. With the rapid development of computer technology and communication technology and the rapid development of visual equipment for the blind, more and more researchers devote themselves to the research of related technology. In this paper, two aspects of dynamic target detection and static object edge detection are studied. The following works are accomplished: firstly, various traditional algorithms and applicable scenes are analyzed, and an improved algorithm suitable for blind walking assistance is adopted. In the aspect of dynamic target detection, ViBe algorithm is used to update the background and count to reinitialize the background model, and the adaptive threshold method is used to extract the edge of the static target. Then the improved algorithm and the traditional algorithm are simulated by MATLAB. The comparison of F_Measure parameters and subjective observation shows that, The recall rate of the improved dynamic target detection algorithm is 10% higher than that of the original algorithm, and the accuracy of the improved algorithm is 6% higher than that of the original algorithm. The improved static target edge detection algorithm can effectively extract interested edge information and eliminate invalid interference. Then the improved algorithm is implemented by Verilog hardware circuit, and the hardware circuit is simulated and analyzed by ModelSim. Finally, the target detection system is set up, which is configured from camera to VGA display, and the processing effect of the whole system is verified on the development board. The result of hardware implementation is compared with the result of MATLAB processing. The maximum relative error between the result of hardware processing and the result of software processing is 4. 3, hardware implementation to meet the design requirements. The experimental results of target detection show that dynamic target detection combined with ViBe algorithm and counting updating background model and counting reinitializing background model strategy can be more suitable for moving scene target detection. The adaptive threshold can reduce the interference of irrelevant information in static target edge detection. The target detection is realized by hardware circuit, which meets the real-time requirement better and enriches the research content of image processing in visual prosthesis.
【学位授予单位】：西安理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN791;TP391.41

【参考文献】