融合超像素与动态图匹配的视频跟踪

发布时间：2018-06-03 08:12

本文选题：目标追踪 + 信息融合　；参考：《西北工业大学学报》2017年01期

【摘要】：针对视频跟踪过程中目标的形变、遮挡、旋转和背景干扰问题,提出一种融合超像素与动态图匹配的视频跟踪方法。首先,采用融合局部熵特征的简单线性迭代聚类(simple linear iterative clustering,SLIC)方法经聚类分析生成超像素集合,使生成的超像素边缘贴合度更好。其次,采用图像分割(graph cuts)方法生成候选目标超像素集合,并融合在线支持向量机学习算法(online SVM learning algorithm,LASVM)分类预测结果,使前景与背景分离的准确度更高。然后,充分利用目标的几何结构信息构建基于图模型的相似度矩阵,解决目标的形变和遮挡问题。理论分析与仿真结果表明:相比现有其他视频跟踪方法,新方法对跟踪过程中的遮挡和形变情况具有较强的鲁棒性,对一定程度的背景干扰和旋转问题跟踪效果良好。
[Abstract]:Aiming at the problems of deformation, occlusion, rotation and background interference in video tracking, a video tracking method combining hyperpixel and dynamic image matching is proposed. Firstly, a simple linear iterative clustering method, is used to generate the super-pixel set by clustering analysis, which makes the edge sticking degree of the generated super-pixel better. Secondly, the image segmentation algorithm is used to generate candidate target super-pixel set, and the online support vector machine learning algorithm online SVM learning algorithm LASVMM-based classification and prediction results are fused to improve the accuracy of foreground and background separation. Then, the similarity matrix based on graph model is constructed by making full use of the geometric structure information of the target to solve the problem of deformation and occlusion of the target. Theoretical analysis and simulation results show that compared with other video tracking methods, the new method is robust to the occlusion and deformation in the tracking process, and has a good tracking effect to some extent of background interference and rotation.
【作者单位】：西北工业大学电子信息学院;光电控制技术重点实验室;
【基金】：光电控制技术重点实验室和航空科学基金(2016515303)资助
【分类号】：TP391.41

【相似文献】