方程式赛车ROS平台下基于TensorRT的YOLOv5算法改进  

Improvement of YOLOv5 algorithm for formula racing based on TensorRT under ROS platform

作　　者：胡天鑫 邓超 马俊杰[1] 刘旺 HU Tianxin;DENG Chao;MA Junjie;LIU Wang(College of Automonile and Traffic Engineering,Wuhan University of Science and Technology,Wuhan 430065,Hubei,China;Intelligent Automobile Engineering Research Institute,Wuhan University of Science and Technology,Wuhan 430065,Hubei,China)

机构地区：[1]武汉科技大学汽车与交通工程学院,湖北省武汉市430065 [2]武汉科技大学智能汽车工程研究院,湖北省武汉市430065

出　　处：《农业装备与车辆工程》2023年第5期14-19,共6页Agricultural Equipment & Vehicle Engineering

基　　金：国家自然科学基金青年科学基金项目:自动驾驶被动脱离状态下驾驶人应急接管认知机理研究(52002298);湖北省自然科学基金计划青年项目:高速公路团雾气候下驾驶应激响应预警机制研究(2020CFB118);湖北省教育厅科学技术研究计划青年人才项目:基于“驾乘意图—车辆状态—动态环境”的碰撞风险综合态势感知方法(Q20201107);教育部产学合作协同育人项目:威士讯自动驾驶产学研协同创新基地(202102580026)。

摘　　要：基于机器视觉的无人驾驶技术是近年来的研究热点,尤其在高速路、路锥道路等特殊路段,机器视觉有明显的优势。针对现有YOLOv4、Faster-RCNN等普通深度学习算法在路锥道路识别速度慢、障碍物识别不稳定等缺陷,基于武汉科技大学无人方程式赛车,提出在ROS平台利用TensorRT加速YOLOv5算法,实现无人方程式赛车环境感知,并接入Gazebo进行赛车的路劲规划仿真。实车实验表明,相比于普通的YOLOv4算法,具有更快的识别速度,一帧图像识别时间在10~20 ms,路锥障碍物的平均识别精度更高,达到99.2%,且识别效果稳定,具有实际应用价值。The unmanned driving technology based on machine vision is a research hotspot in recent years.Especially in special road sections such as highways and road cone,machine vision has obvious advantages.In view of the existing general deep learning algorithms such as YOLOv4,Faster-RCNN,etc.,the recognition speed is slow and obstacles such as cone are unstable under cone road conditions.Based on the unmanned equation car of Wuhan University of Science and Technology,a TensorRT accelerated YOLOv5 algorithm is proposed under the ROS platform to realize the environmental perception of unmanned equation racing cars,and to join Gazebo for the simulation of racing force planning.The real vehicle experiments show that compared with the ordinary YOLOv4 algorithm,the recognition speed is faster,the average recognition accuracy of road cone obstacles is 99.2%higher in a frame of image recognition time of 10~20 ms,and the recognition effect is stable,which has practical application value.

关 键 词：机器视觉 ROS YOLOv5 路锥 

分 类 号：U461[机械工程—车辆工程]