基于OpenVnoi与深度摄像头的实时视障辅助系统设计  

A Real-time Visual-impairment Assistant System Based on OpenVino and DeepSensor Camera

作　　者：梁晓妮 楚朋志 肖雄子彦 冷春涛[1] Liang Xiaoni;Chu Pengzhi;Xiao Xiongziyan;Leng Chuntao(Shanghai Jiao Tong University Student Innovation Center,Shanghai 200240)

机构地区：[1]上海交通大学学生创新中心,上海200240

出　　处：《传动技术》2023年第3期36-40,共5页Drive System Technique

摘　　要：阐述一种基于计算机视觉的视障辅助系统设计方案,该方案硬件基于intel AlxBoard平台进行AI推理,采用intel Realsense D455深度摄像头进行数据采集,应用YoloV5s预训练模型进行目标物体检测,检测出物体位置并根据深度摄像头提供的深度信息计算物体距离,并为视障群体进行语音播报。为提高系统的推理性能,采用OpenVino框架提升推理性能,将推理速度从14 FPS提升至35 FPS,达到实时推理速度,使视障人群使用起来实时、安全。本系统软件基于Python语言及OpenVino库、OpenCV库、Pyttsx3库,通过计算机视觉目标检测方法,实现实时目标检测与物体位置语音提醒功能。该方案未来可通过头盔、眼镜等产品为视障人群服务。A design is described for a computer vision-based visual impairment assistance system.The hardware of this system is based on the Intel AlxBoard platform for AI inference,and uses the Intel Realsense D455 depth camera for data acquisition.The YoloV5s pre-trained model is applied for object detection,which detects the location of objects and calculates the distance between objects based on the depth information provided by the depth camera.Finally,the system provides audio broadcasting for the visually impaired.To improve the inference performance of the system,the OpenVino framework was adopted to increase the inference speed from 14FPS to 35FPS,achieving real-time inference speed and making it safe and practical for visually impaired users.The system software is based on the Python language and the OpenVino library,OpenCV library,and Pyttsx3 library,and uses computer vision target detection methods to achieve real-time target detection and object location voice alerting.This solution can serve visually impaired people through helmets,glasses and other products in the future.

关 键 词：OpenVino 深度摄像头 AlxBoard 目标检测 视障辅助 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]