基于多特征和局部联合稀疏表示的目标跟踪

发布时间：2018-03-17 07:27

本文选题：机器视觉　切入点：目标跟踪　出处：《激光与光电子学进展》2017年10期 　论文类型：期刊论文

【摘要】：针对目标跟踪容易受到遮挡、形变和光照变化影响的问题,在粒子滤波框架下提出一种基于多特征和局部联合稀疏表示的目标跟踪算法。利用HSV空间建立目标的颜色表观模型;利用增强的中心对称局部二值模式建立目标的纹理表观模型,并用局部联合稀疏编码表示。综合颜色和纹理特征计算候选区域与目标的相似性,并利用最大后验概率估计目标当前状态。每2帧判断一次目标表观模型是否需要更新,减少了因频繁更新目标造成的累积误差。利用visual tracker benchmark数据集与其他4种跟踪算法进行了对比实验,结果表明,本文算法的整体精确度和成功率分别为83.5%和79.6%。本文算法在存在遮挡、形变和光照变化的情况下,能够准确稳定地跟踪目标。
[Abstract]:Aiming at the problem that target tracking is easy to be affected by occlusion, deformation and illumination change, a target tracking algorithm based on multi-feature and local joint sparse representation is proposed under the framework of particle filter. The color representation model of the target is established by using HSV space. An enhanced centrosymmetric local binary model is used to build the texture model of the target, and the local joint sparse coding is used to calculate the similarity between the candidate region and the target by combining the color and texture features. Using the maximum posterior probability to estimate the current state of the target, every 2 frames are used to determine whether the target model needs to be updated or not. The cumulative error caused by frequent updating of targets is reduced. The comparison between the visual tracker benchmark dataset and the other four tracking algorithms is carried out, and the results show that, The global accuracy and success rate of this algorithm are 83.5% and 79.6 respectively. The algorithm can track the target accurately and stably under the condition of occlusion deformation and illumination change.
【作者单位】：天津大学电气自动化与信息工程学院;
【基金】：国家自然科学基金(61573060,61673294)
【分类号】：TP391.41

【相似文献】