基于运动模型和数据关联的多目标跟踪  

Multi-object tracking based on motion model and data association

作　　者：侯建华[1] 余宁远 项俊[1] HOU Jianhua;YU Ningyuan;XIANG Jun(College of Electronic Information Engineering,South-Central Minzu University,Wuhan 430074,China)

机构地区：[1]中南民族大学电子信息工程学院,武汉430074

出　　处：《中南民族大学学报（自然科学版）》2023年第2期260-266,共7页Journal of South-Central University for Nationalities：Natural Science Edition

基　　金：国家重点研发计划资助项目(2020YFC1522905)。

摘　　要：在多目标跟踪的拥挤场景中,目标之间的相互遮挡以及目标外观变化,给多目标跟踪中的目标位置预测和数据关联带来了很大的挑战.利用卡尔曼滤波算法建模目标运动模型对目标轨迹进行预测,能够有效缓解目标外观变化的影响.数据关联是多目标跟踪中的重要组成部分,为此,设计了一种相关性网络来处理多目标跟踪中的数据关联.实验结果证明:利用运动模型在跟踪速度上可以实现实时的跟踪效果,设计的相关性网络有效提升了跟踪器的跟踪精度.In the crowded scene of multi-object tracking(MOT),the occlusions among objects and objects appearance changing bring significant challenges to predicting object location and data association in multi-object tracking.It is an effective way to alleviate the impact of appearance changing by using Kalman filter as a motion model to predict the trajectories of targets.A correlation network is designed for data association,a fundamental component in MOT.Experimental results show that the real-time tracking can be achieved with the motion model and the designed correlation network can improve the tracking accuracy effectively.

关 键 词：多目标跟踪 运动模型 数据关联 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]