基于多特征自适应融合的分类采样跟踪算法

发布时间：2018-04-07 16:08

本文选题：视觉跟踪　切入点：粒子滤波(PF)　出处：《光电子·激光》2016年03期

【摘要】：针对目标跟踪中的场景易变和目标模板不稳定等问题,提出了一种基于多特征自适应融合的分类采样跟踪算法。算法利用密集特征信息将目标模板用多个重叠子区域划分,每个子区域对应一个多特征采样窗口。利用多特征自适应融合构造强可区分性的目标模型,最大程度地提高各子区域之间的互补性,以增强目标模板的区分能力。在粒子滤波(PF)框架下,多特征自适应融合策略提高了目标观测质量,保证跟踪的持续稳定。实验结果表明,本文所提算法具有良好的目标跟踪性能,并对动态场景、目标形变及遮挡情况具有较好的跟踪准确性和鲁棒性。
[Abstract]:In order to solve the problem of variable scene and unstable target template in target tracking, a classification sampling tracking algorithm based on multi-feature adaptive fusion is proposed.The algorithm uses dense feature information to divide the target template into multiple overlapped sub-regions, each of which corresponds to a multi-feature sampling window.The multi-feature adaptive fusion is used to construct a highly distinguishable target model to maximize the complementarity among sub-regions and to enhance the discriminative ability of target templates.In the framework of particle filter (PFV), multi-feature adaptive fusion strategy improves the quality of target observation and ensures the stability of tracking.Experimental results show that the proposed algorithm has good target tracking performance and has good tracking accuracy and robustness for dynamic scene, target deformation and occlusion.
【作者单位】：常熟理工学院计算机科学与工程学院;江南大学物联网工程学院;
【基金】：国家自然科学基金(61373055;61300186;41501461) 江苏省自然科学基金(BK20140419) 江苏省高校自然科学研究(14KJB52001) 常熟理工学院科研基金(XZ1309) 苏州大学高校省级重点实验室开放基金(KJS1522)资助项目
【分类号】：TP391.41

【相似文献】