一种基于YOLO深度学习架构的路口交通灯信息车辆间共享方法研究  

A Method of Traffic Light Information Sharing Among Vehicles Based on YOLO Deep Learning Architecture

作　　者：李江天 罗定生[2] LI Jiangtian;LUO Dingsheng(Department of Mechanical Engineering,School of Energy,Power and Mechanical Engineering,North China Electric Power University,Beijing 102206;Department of Machine Intelligence,School of Electronics Engineering and Computer Science,Peking University,Beijing 100871)

机构地区：[1]华北电力大学能源动力与机械工程学院,北京102206 [2]北京大学信息科学技术学院智能科学系,北京100871

出　　处：《系统科学与数学》2022年第2期370-385,共16页Journal of Systems Science and Mathematical Sciences

摘　　要：公交车作为城市交通的重要组成部分,由于其高度明显高于多数车辆,在行驶过程中极易对后方车辆驾驶者观察交通灯的视线造成遮挡.这种遮挡容易造成后方车辆在路口盲目跟随公交车从而导致违规驾驶、车辆追尾等后果.为了解决驾驶视线遮挡问题,文章构建了一种基于YOLO深度学习架构的路口交通灯信息车辆间共享系统,该系统通过摄像设备对路况影像数据进行收集,运用图像预处理手段提高图像的质量,运用基于深度学习架构的目标检测模型对路况影像中是否包含交通灯及交通灯的状态进行实时识别,并将识别结果实时显示在电子屏幕上,从而对后方及周围车辆驾驶者进行提醒,以达到信息共享的目的.为了降低系统对于计算资源的需求,在基于YOLO V3的交通灯识别模型的基础上又开发了基于YOLO Lite的轻量化模型,该模型可以较流畅地运行于CPUi5-750之上.基于实际路况数据在开发板上对系统表现进行了实验验证,证明了此方案的可行性.As a fundamental part of urban traffic,bus is much higher than other kinds of vehicles.Hence,buses may block rear drivers’sights when rear drivers want to observe the traffic lights.During such sight block,the rear drivers tend to blindly follow the bus,which may lead to illegal driving,rear-end collision,etc.To solve this problem,this paper proposes a traffic light information sharing system based on the YOLO deep learning framework.Specifically,the system collects the road image data through the camera,improves the image quality by using the image preprocessing method.The state-of-art Yolo V3 model is utilized to timely identify the position and color of traffic lights within the road image.Finally,the identification results is displayed on the electronic screen to remind the surrounding drivers.To further reduce the demanding of computing resources of Yolo V3,a lightweight model based on Yolo Lite is developed,which can run smoothly on CPU i5-750.The experimental evaluation using real-world road image data proves the effectiveness of the proposed system.

关 键 词：交通灯信息车辆间共享系统 深度学习 YOLO算法 目标检测 图像预处理 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程] U495[交通运输工程—交通运输规划与管理]