摘要
多目标跟踪的研究对于构建人、路、车、云协同一体化的智能交通监控系统具有广泛的应用价值,传统手工设计特征的方法对高层信息的表征能力有限,难以进行复杂场景下的多目标跟踪,随着深度学习的发展,多目标跟踪算法的性能取得较大进展.为了宏观把握基于深度学习的多目标跟踪算法的研究进展,首先比较基于检测的跟踪算法、基于联合检测与跟踪算法、基于单目标跟踪器的多目标跟踪算法的优缺点;然后介绍多目标跟踪算法在智能交通监控场景的应用;最后总结目前多目标跟踪存在的问题与挑战,对多目标跟踪算法未来在智能交通领域的发展进行思考和展望.
To build the integrated intelligent traffic monitoring system based on the cooperation of human,road,vehicle and cloud,the research of multi-object tracking has wide application potentials.Traditional methods with handcrafted features are hard to fully represent high-level information,making it difficult to track multi-targets in complex scenes.Deep learning with its powerful learning ability,has gradually been used in various industries and fields,setting off a wave of smart technologies.To understand the research progress on the multi-object tracking algorithms based on deep learning,firstly,the pros and cons of three tracking algorithms,namely tracking by detection,joint detection and tracking as well as multi-object tracking with single object tracker,are compared.Then,the applications of multi-object tracking algorithm in intelligent traffic monitoring systems are introduced.Finally,the problems and challenges of multi-object tracking algorithm are concluded,and the growing trend of multi-object algorithms in the intelligent transportation field is discussed and forecasted.
作者
金沙沙
龙伟
胡灵犀
王天宇
潘华
蒋林华
JIN Sha-sha;LONG Wei;HU Ling-xi;WANG Tian-yu;PAN Hua;JIANG Lin-hua(School of Information Engineering,Huzhou University,Huzhou 313000,China;Huzhou Institute of Zhejiang University,Huzhou 313000,China)
出处
《控制与决策》
EI
CSCD
北大核心
2023年第4期890-901,共12页
Control and Decision
基金
国家自然科学基金项目(61775139)
浙江省级重点研发计划项目(2020C02020).
关键词
智能交通系统
多目标跟踪
深度学习
智能化
目标检测
研究进展
intelligent transportation systems
multi-object tracking
deep learning
smart technologies
object detection
research progress