基于信息融合的典型工况感知算法研究  

Research on Typical Working Condition Perception Algorithm Based on Information Fusion

作　　者：王梦园 甘海云 袁志宏 WANG Mengyuan;GAN Haiyun;YUAN Zhihong(School of Automotive and Transportation,Tianjin University of Technology and Education,Tianjin 300222;National and Local Joint Engineering Center for Smart Vehicle-Road Collaboration and Safely Technology,Tianjin 300084;College of Energy and Power Engineering,Shandong University,Shandong Jinan 250061)

机构地区：[1]天津职业技术师范大学汽车与交通学院,天津300222 [2]智能车路协同与安全技术国家地方联合工程研究中心,天津300084 [3]山东大学能源与动力工程学院,山东济南250061

出　　处：《汽车实用技术》2021年第22期17-23,共7页Automobile Applied Technology

基　　金：基于封闭园区及开放道路的L4级智能网联汽车研发及示范运行(编号18ZXZNGX00230)。

摘　　要：针对自动驾驶频发交通事故的问题,结合中国道路的典型工况,文章基于信息融合的理念,将毫米波雷达与摄像头融合的目标检测的结果进行融合来解决盲区中行人横向穿越道路的小目标检测问题。首先将毫米波雷达和摄像头环境感知的信息进行融合;然后通过yolov2目标检测算法对相邻车道前方车辆进行检测;最后,对车辆前方划定ROI(Region of Interest),并通过yolov3-bt对ROI进行检测。实验对比结果表明,对车辆前方出现行人这一现象,毫米波雷达与摄像头信息融合的方法比单视觉算法检测提前135帧,即检测时间提前4.5 s,提升了17.8%。表明文章提出的毫米波雷达与摄像头信息融合的方法可以进一步提高自动驾驶车辆行驶的安全性。Aiming at the frequent traffic accidents of autonomous driving,combined with the typical working conditions of Chinese roads,an information fusion scheme is proposed in this paper.Firstly,the millimeter-wave radar and camera are used to perceive the front environment,and the collected information is fused.Secondly,the front right vehicle is detected based on the yolov2 target detection algorithm,which forms the trigger condition for the appearance of a specific scene.Finally,the area of interest in front of the vehicle is divided,and the yolov3-bt algorithm is used for target detection in the area of interest to solve the problem of small target detection under typical working conditions.Through experimental comparison,it is found that the fusion algorithm improves the detection time by 17.8%compared with the ordinary vision detection algorithm.Aiming at the phenomenon of pedestrians in front of the vehicle,compared with ordinary visual detection algorithms,the improved algorithm is advanced by 135 frames on average,and the corresponding detection time is advanced by 4.5s,thereby further improving the safety of intelligent driving vehicles.

关 键 词：自动驾驶 环境感知 信息融合 目标检测 

分 类 号：U495[交通运输工程—交通运输规划与管理]