摘要
单摄像机视觉跟踪过程中,常发生目标被遮挡或背景复杂的情况,此时容易跟丢目标,为了提高跟踪的准确性。从目标表现和背景的不确定性入手,以协方差特征对目标表现以及背景进行建模,应用到到粒子滤波的框架中,优化采样粒子的分布,在估计粒子的权重时,不仅考虑目标的真实状态和可能状态的相似性,还考虑了目标可能的状态和背景的差异.将提出的算法与粒子滤波,均值漂移,基于协方差概率跟踪算法进行比较,通过MATLAB2010编程平台,比较了几种算法的处理速度以及跟踪误差,试验结果表明,提出的算法每秒处理速度为60帧/s,优于上述3种跟踪算法平均误差值也高于另外3种算法。所提出算法在目标存在遮挡和背景较为复杂时,能够保证对目标进行准确,连续的跟踪。
Case of a single camera visual tracking process often occurs goal is blocked or complex background,this time with the lost easy target,in order to improve the accuracy of tracking The purpose of this paper,the performance from the target and the background of uncertainty start to co- variance performance characteristics of the target and the background modeling,while the sequential Monte Carlo filter proposed framework as applied to visual target tracking to track and optimize the distribution of the sample particles,heavy particles in the right estimate,not only to consider the true state of the target and may similarity of the state,but also consider the differences between the target and the background of possible states: The proposed algorithm and particle filter,mean shift tracking algorithm based on the probability of covariance comparing programming platform via MATLAB 2010 compare several the processing speed of the algorithm and the tracking error,the test results show that the proposed algorithm per second processing speed of 60 frames / sec,better than the average of the three tracking algorithm error value is also higher than the other three algorithms. This paper presents algorithms target occlusion exists when the background is complex,the target to ensure accurate and continuous tracking.
出处
《电子测量与仪器学报》
CSCD
北大核心
2015年第2期289-295,共7页
Journal of Electronic Measurement and Instrumentation
关键词
拟蒙特卡洛滤波
目标跟踪
协方差
Quasi-Monte Carlo filtering
object tracking
covariance