一种多特征融合的车辆追踪算法的研究与实现  

Research and Implementation of Vehicle Tracking Algorithm Based on Multi-feature Fusion

作　　者：刘慧敏 关亨 于明鹤 赵志滨 LIU Hui-min;GUAN Heng;YU Ming-he;ZHAO Zhi-bin(School of Computer Science and Engineering,Northeastern University,Shenyang 110819,China;School of Software,Northeastern University,Shenyang 110819,China)

机构地区：[1]东北大学计算机科学与工程学院,沈阳110819 [2]东北大学软件学院,沈阳110819

出　　处：《小型微型计算机系统》2020年第6期1258-1262,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金重点项目(U1435216)资助.

摘　　要：多目标追踪的主要任务是在给定图像序列中,获取每帧图像中目标的信息,并将图像序列中的目标关联起来.在高速公路的视频检测中,需要对视频序列中检测出来的车辆进行追踪,它是车流量统计、异常事件检测等工作的基础.本文提出一种简化高效的多目标追踪方法,在关联运动物体位置、交并比(IOU)的基础上,增加了颜色特征,并且使用无损卡尔曼滤波进行位置修正.这种追踪方法可以有效地处理目标遮挡、丢失问题.算法实现简便、运算速度快,可以满足实时处理的需求.本文采用高速公路的视频录像作为数据集进行实验,实验结果表明本文方法可以有效地处理车辆追踪问题.The main task of multi-object tracking is to associate targets in diverse images by detected information from each frame of a given image sequence.For the scenario of highway video surveillance,the equivalent research issue is vehicles tracking,which is necessary and fundamental for traffic statistics,abnormal events detection,traffic control et al.In this paper,a simplified and efficient multi-object tracking strategy is proposed.Based on the position and intersection-over-union(IOU)of the moving object,the color feature is derived,and unscented Kalman filter is involved to revise targets’positions.This innovative tracking method can effectively solve the problem of target occlusion and loss.The simplicity and efficiency makes this algorithm applicable for the perspective of realtime system.In this paper,highway video recordings are explored as data repository for experiments.The results show that our method outperforms on the issue of vehicle tracking.

关 键 词：视频监控 位置预测 多目标追踪 数据关联 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]