摘要
要为解决导盲眼镜穿戴体验感差、反馈效率较低及无法有效帮助盲人过马路等问题,本文以云平台为核心,融合机器视觉、图像处理、无线通信等技术,设计了一款基于机器视觉的智能导盲系统。系统由眼镜端、云服务端、手机APP端三大部分组成。眼镜端以搭载Linux系统的树莓派及高清双目模组作为主控制器和图像采集设备。视觉深度学习处理云服务端由云服务器,场景分析和语义分割等构成。总运算延迟在4-6s以内,盲人在独立出行时能及时得到反馈信息。本研究基于Yolov5识别算法对障碍物进行识别,融合了测距、HSV向量空间、语义分割等帮助盲人通过听觉获得出行场景信息,保证盲人在出行方面的避障安全性和导航便利性,实现智能辅助盲人出行的功能。
In order to solve problems such as poor wearing experience,low feedback efficiency and in-ability to effectively help blind people cross the road,this paper takes cloud platform as the core and in-tegrates machine vision,image processing,wireless communication and other technologies to design a cloud platform based vision guide system.The system consists of three parts:glasses terminal,cloud service terminal and mobile APP terminal.At the glasses end,Raspberry PI and HD binocular module equipped with Linux system are used as the main controller and image acquisition equipment.The cloud service terminal of visual deep learning processing is composed of cloud server,scene analysis and road crossing evaluation model.The total operation delay is within 0.6-0.9s,and the blind can get timely feedback when traveling independently.This project uses Yolov5 model as the central point to identify obstacles,and integrates ranging,HSV vector space,semantic segmentation and other algorithms to help blind people get travel scene information through hearing,basically ensure the safety of obstacle avoid-ance and navigation convenience for blind people in travel,and realize the function of intelligent assis-tance for blind people to travel.
作者
刘少杰
李建华
毛润好
匡钟慧
杨洪钦
Shaojie Liu;Jianhua Li;Runhao Mao;Zhonghui Kuang;Hongqin Yang(College of Photonic and Electronic Engineering,Fujian Provincial Key Laboratory of Photonics Technology,Fujian Normal University,Fuzhou 350117;Ctff Information Technology Co.,Ltd.,Fuou 350003)
出处
《福光技术》
2023年第1期38-45,共8页
FUJIAN OPTICAL TECHNOLOGY
基金
福建省自然基金重点项目(2021J02028)
大学生创新创业训练计划项目(CXXL-2022205).
关键词
智能导盲
机器视觉
Yolov5识别
双目立体匹配算法
语义分割
Intelligent guide
Machine vision
Yolov5 recognition
Binocular stereo matching algo-rithm
Semantic segmentation