基于YOLOv3算法的盲道障碍物识别技术应用研究  

Research on the Application of Blind Track Obstacle Recognition Technology Based on YOLOv3 Algorithm

作　　者：孙嘉鑫 吴琳 SUN Jia-xin;WU Lin(College of Computer Science,North China University of Technology,Beijing 100144)

机构地区：[1]北方工业大学信息工程学院,北京100144

出　　处：《现代计算机》2021年第5期71-74,共4页Modern Computer

基　　金：大学生科研训练项目(No.218051360020XN214)。

摘　　要：因大风、人为等因素使盲道上存在不同障碍物,从而影响到盲人的出行,通过当前比较流行的实时检测算法——YOLOv3,可对盲道上的障碍物进行识别。可通过OpenCV将盲道区域进行提取,后将处理过的图片通过YOLO检测模型进行物体识别,通过多次不同的图片进行训练、调整参数得到最佳的检测模型。实验结果表明,经过处理后的图片障碍物的平均准确率为96%,检测速度可达15fps,从而在一定程度上为系统应用提供基础。There are different obstacles on the blind road due to strong wind and man-made factors,which affect the travel of blind people.Through the current popular real-time detection algorithm,YOLOv3,the obstacles on the blind road can be identified.The blind path area can be extracted by OpenCV,and then the processed images can be recognized by Yolo detection model.The best detection model can be obtained by training and adjusting parameters of different images for many times.The experimental results show that the average accuracy of the pro⁃cessed image obstacles is 90%,and the detection speed can reach 15 fps,which provides the basis for the system application to a certain extent.

关 键 词：YOLOv3 目标检测 图像识别 盲人出行 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]