高空远距离多目标视觉跟踪模型优化仿真  

Optimization and Simulation of Visual Tracking Model of High-Altitude Long Distance and Multiple Target

作　　者：姚楠 蔡越 余劲[3] 

机构地区：[1]江苏省电力公司电力科学研究院,江苏镇江212000 [2]南京音视软件有限公司,江苏南京210000 [3]南京大学计算机科学与技术学院,江苏南京210000

出　　处：《计算机仿真》2014年第8期427-431,共5页Computer Simulation

摘　　要：在智能跟踪建模优化问题的研究中,高空远距离的视觉目标跟踪一直是智能跟踪领域的难点。因为跟踪目标的距离较远,可追踪特征在色彩、像素灰度等方面会发生较大幅度的衰退和丢失。传统的目标视觉跟踪方法在上述情况下,会迅速丧失跟踪能力,造成目标跟踪丢失。主要因为传统算法在视频目标跟踪算法中未考虑到特征丢失带来的先验的目标信息的问题。提出一种新的目标跟踪模型。模型在跟踪过程中,像素分布的产生直接采用先验概率。引入改进的灰预测GM1模型,通过灰预测改进的GM1的预测值来产生新的建议分布,使得后验概率分布更加逼近真实目标的后验概率密度,保证弱化跟踪的强关联性。实验结果表明,与标准的粒子滤波算法进行对比试验,所提出的算法在远程高空视觉目标跟踪中具有更好的性能。In the study of optimization problem of intelligent tracking modeling, high altitude and long distance visual target tracking has been the difficulty in the field of intelligent tracking. Because the distanceof tracking target is far, the traceable characteristics in terms of color, gray level of pixel 1 and so on decline and lose significantly. The traditional target visual tracking method, in this case, will quickly lose track ability, causing the loss of target tracking. It mainly because the problem of transcendental target information caused by the characteristics lost is not considered in the traditional algorithm in visual target tracking algorithm. A new model of target tracking is proposed in this paper. Model in the process of tracking, the prior probability is directly used in the generation of the distribu- tion of the pixels. The improved grey level prediction model of GM1 is introduced, through grey level predict the predicted value of improvedGM1 this forecast to generate a new proposal distribution, making that the posterior probabili- ty distribution is more close to the posterior probability density of the real target, which can guarantee the strong rele- vancy of weakening tracking. The experimental results show that compared to t the standard particle filter algorithm, the proposed algorithm used in long - distance and high altitude visual target tracking has better performance.

关 键 词：目标跟踪 粒子滤波 建议分布 

分 类 号：TP123[自动化与计算机技术—控制理论与控制工程]