典型复杂背景下基于变分的目标轮廓跟踪算法及实现技术研究

项目名称： 典型复杂背景下基于变分的目标轮廓跟踪算法及实现技术研究

项目编号： No.61473310

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 其他

项目作者： 周则明

作者单位： 中国人民解放军国防科技大学

项目金额： 82万元

中文摘要： 研究典型复杂背景下基于变分的目标轮廓跟踪模型。在变分水平集跟踪框架中引入先验知识，能够显著改善低对比度、部分遮挡、杂乱背景下目标的跟踪性能。研究运动目标先验形状与先验光谱知识的表达方法，抽取其线性及非线性变化模式，提出带先验知识约束能量项的变分跟踪模型；分析跟踪算法的快速数值计算方法；基于仿射变换参数预测运动目标的位置并估计其运动状态。稀疏表示方法能够提高模型描述目标非线性变化的能力；在线学习算法能够提高跟踪模型的鲁棒性；将跟踪范围限定在上一帧跟踪轮廓的窄带中提高了模型的计算效率。以低空中小目标慢速运动的视频、心脏在一个心动周期中的核磁共振图像序列等为测试数据，验证刚体和弹性体目标跟踪算法的有效性。该项目的成果能够应用于刚性目标及医学影像序列的分析，为评估目标的运动状态提供必要的理论依据。

中文关键词： 目标跟踪；变分方法；水平集；在线学习；稀疏表示

英文摘要： Contour tracking models of the object in the typical complex background based on variational technique are studied in this proposal. The prior knowledge of the object is introduced into the variational level set framework, which remarkably improves the tracking performance on the condition of low-contrast, part occlusion and clutter environments. The research topics mainly include: resprensentation of the prior shape and spectrum information; analysis of the linear and non-linear variation mode; variational tracking model with prior knowledge constraints; efficient numerical computation approaches of the tracking algorithm and estimation and prediction models of the object location and the motion state based on the affine transform parameters. Sparse representation of the prior shape can capture more details of the non-linear variation mode while online learning algorithm for updating variation pattern space can improve the robustness of the tracking model. The tracking area is set in the narrow band around the object contour in the previous frame, which accelarates the numerical computation.The effectiveness of the tracking algorithms for the rigid and deformable objects is validated on the datasets including the videos of the moving object in the low-altitude areaspace at a low speed and the cardiac MRI sequences in a cycle.The results could be applied in the analysis of the little rigid object and the medical imagery sequences,which provides the guidance for the assessment of the motion state.

英文关键词： object tracking;variational approach;level set;online learning;sparse representation