基于相关滤波器的自适应目标跟踪

发布时间：2018-10-09 13:59

【摘要】：针对传统核相关滤波器(KCF)跟踪算法在遮挡情况下通常会跟踪失败问题,提出一种目标边缘增强的自适应核相关滤波器算法。首先通过核相关滤波器找出目标位置,对目标位置区域进行边缘增强,使目标特征更突出,提高分类器性能,然后通过计算目标位置的响应强度自适应的改变模板的学习率参数,使检测模板适应外界环境的变化。实验选取15段公开视频序列进行测试,与现有几种相关滤波器进行比较,相对于结果最好的KCF算法的平均中心位置误差减少了9.6像素,平均成功率提高了7.55%,平均距离精度提高了21.62%。实验结果表明在目标被遮挡、旋转及快速运动等复杂情况下,该算法有较强的适应性,具有重要的研究和应用价值。
[Abstract]:In order to solve the problem of tracking failure in the case of occlusion, an adaptive Kernel correlation filter (KCF) algorithm with edge enhancement is proposed for the traditional kernel correlation filter (KCF) tracking algorithm. Firstly, the target location is found by the kernel correlation filter, and the edge of the target location region is enhanced, which makes the target feature more prominent and improves the performance of the classifier. Then, the learning rate parameters of the template are adaptively changed by calculating the response strength of the target position, so that the detection template adapts to the change of the external environment. The experiment selected 15 segments of open video sequences to test, compared with several existing correlation filters, compared with the best result of the KCF algorithm, the average center position error is reduced by 9.6 pixels. The average success rate increased by 7.55 and the average distance accuracy increased by 21.62. The experimental results show that the algorithm has strong adaptability and has important research and application value in complex cases such as object occlusion, rotation and fast motion.
【作者单位】：中国石油大学(华东)计算机与通信工程学院;中国科学院计算技术研究所智能信息处理重点实验室;
【基金】：国家自然科学基金项目(61379106) 山东省中青年科学家奖励基金项目(BS2010DX037) 山东省自然科学基金项目(ZR2009GL014,ZR2013FM036,ZR2015FM011) 浙江大学CAD&CG国家重点实验室开放课题(A1315) 中央高校基本科研基金项目(13CX06007A,14CX06010A,14CX06012A)
【分类号】：TP391.41

【相似文献】