文档介绍：华侨大学硕士论文
摘要
在众多的人脸跟踪算法中,Mean Shift(均值偏移)算法作为一
种无参数密度估计的高效的模式匹配算法,很容易作为一个模块和其
它算法集成,在跟踪领域已得到广泛的研究和应用。
但是Mean Shift算法本身也存在一些不足,在基于Mean Shift算法
的人脸跟踪系统中,它只利用了人脸的肤色信息,没有利用目标的运
动信息,当出现遮挡时,容易丢失目标,导致跟踪失败。本文针对这
些不足,提出两种改进的算法:
本文首先提出一种Mean Shift和Kalman滤波相结合的算法。
Kalman滤波预测目标在本帧的可能位置,Mean Shift算法在该点邻域
内搜索,减小了搜索区域,提高了系统的实时性。实验结果表明,该
算法对快速运动目标的跟踪效果良好,而且能较好地解决遮挡引起的
目标丢失问题。
Kalman滤波是线性预测器,无法预测复杂运动的目标。为此,本
文提出一种Mean Shift和粒子滤波相结合的算法,利用Mean Shift算法
把重采样后的粒子收敛到可能的目标位置。由于每个粒子表示状态更
合理,所需粒子数大大减少,算法的实时性得到了提高。实验结果表
明,改进算法在复杂环境下目标快速晃动和出现遮挡时都能够实现很
好的跟踪,相对于传统粒子滤波器实时性得到很大提高。
关键词:人脸跟踪均值偏移卡尔曼滤波粒子滤波
II
华侨大学硕士论文
Abstract
Among many methods about face tracking technology, Mean Shift is
a nonparametric estimation and high effective pattern matching algorithm.
Since the algorithm can be bined with other algorithm to build
good tracker, it has been extensively studied and widely used in the object
tracking.
The ing of a Mean Shift based face tracking system is it
uses only plexion information but no motion information. When
the occlusion occurs, the object is easily lost resulting in the tracking
failure. In order to improve the tracking ability, two improved mean shift
algorithms are proposed in this thesis:
The first one is Mean-bined with Kalman filter tracking
algorithm, where a Kalman filter is employed to predict the object
position in current frame. As a result, the searching area of the
Mean-Shift is diminished and the tracking speed is improved. The
Experiment results show that the proposed algorithm can not only track
the fast moving object well, but also deal with the occlusion efficiently.
Because Kalman filter is a linear predictor, it cannot be used in
complex motion prediction. Therefore, another face tracking algorithm,
particle filter based mean shift face tracking scheme is proposed. The
Mean-Shift algorithm converge re-sampling particles to candidate areas
III
知识水坝***@pologoogle为您整理
华侨大学硕士论文
of