基于机器视觉的导盲杖设计

Design of Guide Rod Based on Machine Vision

【目的】为了满足视障人士多方面的实际出行需求,设计了一款基于机器视觉的导盲杖。【方法】该系统整体为模块化设计,采用了STM32F407VET6单片机和边缘智能计算K210芯片为核心。其机器视觉功能是利用神经网络和YOLOv2算法训练建立起模型,并对摄像头拍摄到的实时图像进行检测而实现的。系统自身具备陀螺仪姿态角度测量功能,通过判断系统旋转角度是否达到设定阈值,以此来判断使用者是否有摔倒迹象;系统还具有GPS定位功能,在突发情况下系统会通过串口发送AT指令控制GSM模块向紧急联络人发送位置信息。【结果】该系统各个功能经模拟测试,其机器视觉功能检测的正确率达到99%以上;姿态角度测量误差维持在1°以内;GPS和GSM功能可以定位并发送短信息。【结论】该系统的功能可满足视障人士基本的实际出行需要。此外,图像检测模型的建立,可为今后训练模型以检测更多物体,完善机器视觉功能,应对更多实际出行的情况。

[Purposes]In order to meet the practical travel needs of blind people in various aspects,a ma⁃chine vision based guide stick has been designed.[Methods]The overall design of this system is modu⁃lar,with the core using STM32F407VET6 microcontroller and edge intelligent computing K210 chip;its machine vision function is achieved by using neural networks and YOLOv2 algorithm to train and estab⁃lish models,and detecting real-time images captured by cameras.The system itself has a gyroscope atti⁃tude angle measurement function,which determines whether the system rotation angle has reached the set threshold to determine whether the user has any signs of falling;the system also has GPS positioning function,and in case of emergencies,the system will send AT commands through the serial port to con⁃trol the GSM module to send location information to the emergency contact person.[Findings]All func⁃tions of this system have been simulated and tested,and the accuracy of machine vision function detec⁃tion can reach over 99%;the attitude angle measurement error is maintained within 1°;GPS and GSM functions can locate and send short messages.[Conclusions]The above functions can meet the basic practical travel needs of blind people.In addition,with the establishment of image detection models,fu⁃ture models can be trained to detect more objects,thereby improving machine vision functions and re⁃sponding to more actual travel situations.