摘要
为了解决出租车与黄色小型车辆外观相似、不易区分的问题,以哈尔滨市出租车为研究对象,以YOLOv5+DeepSort为基本框架,新增交通量与速度检测模块。基于视频采集数据,建立出租车目标检测数据集与出租车图像数据集,采用深度学习方法构建车型识别模型。建立了考虑出租车比例因素的速度影响模型,分析了畅行状态下出租车运行特征。结果表明:结合深度学习的出租车车型识别精确率高达0.88;畅行状态下出租车平均速度比其他车型高5~15 km/h;出租车比例对全局平均速度及速度-流量曲线增长趋势存在一定影响;考虑出租车比例的速度影响模型在继承传统BPR模型优点的同时,精度提升了20%左右。
To address the issue of taxis and small yellow vehicles appearing similar and being difficult to distin⁃guish,this study focuses on the taxis of Harbin city.Using YOLOv5+DeepSort as the foundational framework,a new module for traffic volume and speed detection was constructed.Based on video collection data,a dataset for taxi target detection and a taxi image dataset were established.Deep learning methods were employed to construct a vehicle model identification model.A speed impact model that considers the proportion of taxis was developed,analyzing the operational characteristics of taxis under smooth traffic conditions.The results indi⁃cate that the accuracy of taxi model identification using deep learning reaches up to 0.88,the average speed of taxis in smooth traffic conditions is 5-15 km/h higher than other vehicle types,and the proportion of taxis has a certain influence on the overall average speed and the growth trend of the speed-flow curve.The speed im⁃pact model that considers the proportion of taxis improves precision by about 20%while inheriting the advanta⁃ges of the traditional BPR model.
作者
徐慧智
陈爽
刘嘉玲
蒋时森
陈祎楠
XU Huizhi;CHEN Shuang;LIU Jialing;JIANG Shisen;CHEN Yinan(School of Civil and Transportation Engineering,Northeast Forestry University,Harbin 150040,China;School of Mecha-tronic Engineering,Northeast Forestry University,Harbin 150040,China)
出处
《大连交通大学学报》
CAS
2024年第5期33-41,共9页
Journal of Dalian Jiaotong University
基金
国家自然科学基金青年科学基金项目(71701041)
黑龙江省自然科学基金项目(LH2019E007)。
关键词
交通运输规划与管理
深度学习
出租车
运行特征
车型识别
transportation planning
management
deep learning
taxi
operating characteristics
vehicle type identification
