光学学报, 2018, 38 (2): 0215002, 网络出版: 2018-08-30   

基于在线检测和尺度自适应的相关滤波跟踪 下载: 1152次

Correlation Filter Tracking Based on Online Detection and Scale-Adaption
作者单位
国家数字交换系统工程技术研究中心, 河南 郑州 450000
摘要
针对相关滤波跟踪在遮挡及目标尺度变化等情况下容易跟踪失败的问题,提出一种基于在线检测和尺度自适应的相关滤波跟踪算法。相关滤波跟踪器融合方向梯度直方图特征、颜色属性特征和光照不变特征进行目标定位;通过局部稀疏表示模型的重构残差进行遮挡判别,如果发生遮挡则进行在线支持向量机检测,实现目标重定位;进行由粗至精的尺度估计,通过尺度预估计和牛顿迭代法得到目标的精确尺度。采用均衡的模型更新策略,固定更新相关滤波器,保守更新稀疏表示模型和支持向量机。实验结果表明:与现有跟踪算法相比,所提算法能有效降低遮挡、目标尺度变化等复杂因素的干扰,并在50组测试序列上取得较高的距离精度和成功率,其整体性能优于其他对比算法。
Abstract
In correlation filter tracking, occlusion and object scale change can lead to tracking failure easily. To deal with this problem, a correlation filter tracking algorithm based on online detection and scale-adaption is proposed. The target is initially located through a correlation filter tracker fusing histogram features of oriented gradient, color attribute features and illumination invariant features. The reconstruction residual of local sparse representation model is used for occlusion discrimination. If occlusion occurs, online support vector machine detection will be carried out and target relocating will be realized. Scale estimation from coarse to precise is carried out, and precise scale of target is obtained by scale pre-estimation and Newton iterative method. A balanced model updating strategy is used to update correlation filter regularly and update sparse representation model and support vector machine conservatively. Experimental results show that, compared with existing tracking algorithms, the proposed algorithm can effectively reduce the occlusion, target scale change and other complicated factors, and can gain higher distance precision and success rate on 50 groups of test sequences. The overall performance of the proposed algorithm is better than other contrast algorithms.

王艳川, 黄海, 李邵梅, 高超. 基于在线检测和尺度自适应的相关滤波跟踪[J]. 光学学报, 2018, 38(2): 0215002. Yanchuan Wang, Hai Huang, Shaomei Li, Chao Gao. Correlation Filter Tracking Based on Online Detection and Scale-Adaption[J]. Acta Optica Sinica, 2018, 38(2): 0215002.

本文已被 7 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!