| 引用本文: | 赵运基,裴海龙.子空间描述的关节式目标跟踪[J].控制理论与应用,2013,30(5):604~610.[点击复制] |
| ZHAO Yun-ji,PEI Hai-long.Articulate object tracking based on subspace representation[J].Control Theory and Technology,2013,30(5):604~610.[点击复制] |
|
| 子空间描述的关节式目标跟踪 |
| Articulate object tracking based on subspace representation |
| 摘要点击 2700 全文点击 2096 投稿时间：2011-10-25 修订日期：2013-01-18 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/CTA.2013.11204 |
| 2013,30(5):604-610 |
| 中文关键词 奇异值分解 子空间描述 粒子滤波 局部二值模式 |
| 英文关键词 singular value decomposition subspace representation particle filter local binary pattern |
| 基金项目 国家自然科学基金重点资助项目(61174053, 61104083); 高等学校博士学科点专项科研基金资助项目(20100172110023). |
|
| 中文摘要 |
| 针对关节式目标变化对子空间描述造成的影响, 本文提出了一种基于增量学习的关节式目标跟踪算法. 该算法通过引入图像分割方法与快速傅里叶变换可有效消除背景像素对目标描述造成的影响以及目标区域前景目标位置对不准造成的误差, 同时应用局部二值模式增加目标描述中像素点间的几何位置信息, 应用基于增量学习的方法实现目标特征的在线更新, 最终为跟踪算法提供较为精确的目标描述. 实验结果表明, 本文提出的关节式目标跟踪算法具有较好的目标跟踪效果. |
| 英文摘要 |
| Subspace representation of articulate object is subject to its variation. In order to deal with this problem, an articulate object tracking algorithm based on incremental learning is proposed. In this algorithm, graph-cut algorithm is used to reduce the influence of the background pixels on the object representation. Fast Fourier transform (FFT) is used to reduce the error of matching between the foreground locations in object regions. Local binary pattern (LBP) can increase spatial location information of pixels in the object representation. The object features updated online by incremental learning method can provide more accurate object representation. Experimental results demonstrate that the algorithm is able to track articulated objects with higher accuracy. |