摘要
为了提高对机动目标跟踪的实时性,提出了一种将均值漂移嵌入高斯-厄米特粒子滤波器的目标跟踪算法.通过粒子滤波产生一组带权粒子,在高斯-厄米特预测的基础上利用基于颜色直方图分布的均值漂移算法对各粒子进行迭代优化,由于在提高粒子质量的同时有效降低了维持"多峰"假设所需的粒子数,从而保证了算法的精度和效率.实验结果表明文中算法在保持较高精度的同时,大大提高了跟踪的实时性.
A new algorithm which embeds mean shift algorithm into the Gauss-Hermite particle filter is proposed to improve the execution speed of target tracking system. A set of partieles with weights are created and predicted with Gauss-Hermite filter, and then optimized by shifting to their neighboring modes with mean shift iteration based on color histogram. Fewer particles are needed to maintain multiple hypotheses, consequently improve both accuracy and efficiency. Contrastive experiments between this algorithm and both the traditional mean shift and Gauss-Hermite particle filter are made with two video sequences. The experimental results show that the algorithm enhances tracking efficiency greatly and remains the same accuracy.
出处
《微电子学与计算机》
CSCD
北大核心
2009年第10期61-64,共4页
Microelectronics & Computer
