一种改进的Mean-Shift实时车辆跟踪算法  

Improved Mean-Shift Algorithm of Real-time Vehicle Tracking

作　　者：张彤[1] 马延柯 首照宇[2] 

机构地区：[1]桂林电子科技大学机电工程学院,广西桂林541004 [2]桂林电子科技大学认知无线电与信息处理省部共建教育部重点实验室,广西桂林541004

出　　处：《电视技术》2015年第17期126-130,共5页Video Engineering

基　　金：国家自然科学基金项目(61362021);广西自然科学基金项目(2013GXNSFDA019030;2013GXNSFAA019331;2012GXNSFBA053014;2012GXNSFAA053231;2012jj AAG0080);广西科技开发项目(桂科攻1348020-6;桂科能1298025-7);广西教育厅项目(201202ZD044;2013YB091);桂林市科技攻关项目(20130105-6;20140103-5)

摘　　要：针对传统Mean-Shift算法存在无法有效的估计目标方向及变速运动目标这一缺陷,提出了一种改进的Mean-Shift自适应跟踪算法,在传统算法基础上增添了next帧目标预测功能,并应用于智能交通系统。该算法首先提取运动历史信息,并建立数学模型,据预测理论对next帧目标位置进行预测,使其作为Mean-Shift迭代初始,提高迭代效率。实验表明,该跟踪算法具有良好的实时性和鲁棒性,受车辆方向和速度变化影响较小。Mean-Shift tracking algorithm is widely used in the field of video surveillance, it is a classical method. But there are lar- ger defects for tracking the changing of direction and speed target under the situation of complex movements, In view of this prob- lem, in this paper, an improved method based on Mean-Shift adaptive tracking is proposed to overcome this disadvantage. The function of prediction next target is added in this method, and the method is also used in intelligent transportation systems The al- gorithm takes the position of next target predicted as the Mean-Shift iteration initial position, in order to reduce their number of it- erations, and rapidly converges to the current position. This method is verified that it has real-time, robustness and less influenced by the changing of direction and speed.

关 键 词：Mean—Shift 运动历史信息 目标预测 车辆跟踪 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]