几何布局和行为检测相结合的遮挡目标跟踪  

Tracking objects with occlusion based on geometry layout and activity detection

作　　者：吴雪刚[1,2] 唐远炎[1,3] 房斌[1] 赵爽文 李柳柏[2] 谭勇[2] 邢昌元[1,2] 

机构地区：[1]重庆大学计算机学院,重庆400044 [2]长江师范学院数学与计算机系,重庆408100 [3]澳门大学计算机信息科学系 [4]涪陵区经济和信息化委员会,重庆408000

出　　处：《计算机工程与应用》2015年第13期42-46,共5页Computer Engineering and Applications

基　　金：重庆市教委项目(No.KJ111302)

摘　　要：提出一种基于本地环境几何结构信息并结合行为识别的粒子滤波器来解决行进中被跟踪目标的遮挡问题。当发生遮挡情况时,先测定目标在环境中的相对位置信息,根据先验知识确定此位置目标的相对高度,进一步确定目标地面着力点的位置坐标;根据着力点的位置确定不同目标分属的相对不同的环境区域;再利用行为识别算法的对不同行为(跑、跳、走等)进行判断;将这些参数结合到粒子滤波器中进行预测跟踪。此时,不同位置区域目标在不同轨道上的粒子权重就会增加,不同特征的行为在运动过程的粒子权重也会增加,从而提高了遮挡情况下目标跟踪的准确率。仿真实验结果表明,该方法具有较好的实用性和研究价值。Based on geometry layout and particle filter,this paper proposes a method to solve the problem of object tracking.Prior knowledge from interesting object and its perspective effect,the region of different shapes is investigated.When occlusion happens between different objects,it firstly measures the distance between camera and object,and then gets the relative height of object,further ensures the accurate site of the object.At last,it is known that different objects belong to their regions.And then,using particle filter to tracking different occlusion objects,the weight will be increased when objects in different region move to their rails,the accurate rate will be advanced.The experimental results show the method is with practical effect.

关 键 词：目标跟踪 卡尔曼滤波器 粒子滤波 粒子退化 遮挡 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]