| 摘 要: 为了提高在复杂背景下进行跟踪的精度,在KCF(Kernel Correlation Filter)算法的基础上提出了一种改进方案。首先,提取HOG(Histogram of Gradient)、CN(Color-Naming)和LBP(Local Binary Pattern)三种特征进行融合,获得充分的目标特征信息;其次,引入尺度滤波机制用于估计目标最佳尺度大小,得出最合适的跟踪框;最后,提出了一个峰值更新策略,确保模型更新不受错误信息干扰。实验表明,改进后的算法比KCF算法在精确度和成功率上分别提升了6.5%和4.8%,并且在处理尺度变化、变形、旋转等方面也有很好的鲁棒性。 |
| 关键词: 核相关滤波;目标跟踪;尺度自适应;多特征融合 |
|
中图分类号: TP391
文献标识码: A
|
|
| Kernel Correlation Filter Algorithm Based on Multi Feature Fusion and Scale Adaptation |
|
ZHENG Huiguo, WANG Zhuangzhuang, CAI Zhiming
|
(Fujian University of Technology, Fuzhou 350118, China)
1196462180@qq.com; 1582765508@qq.com; caizm@fjut.edu.cn
|
| Abstract: This paper proposes an improved algorithm based on KCF(Kernel Correlation Filter) algorithm to enhance the tracking accuracy in intricate background. Firstly, three features, HOG(Histogram of Gradient), CN(Colornaming) and LBP(Local Binary Pattern), are extracted for fusion to acquire adequate target feature data. Secondly, a scale filtering mechanism is introduced to estimate the optimal scale size of the target and obtain the most suitable tracking frame. Finally, a peak update strategy is proposed to ensure that model updates are not affected by error information interference. The experiment results show that the improved algorithm has 6.5% higher accuracy and 4.8% higher success rate than the KCF algorithm, and it also shows good robustness in handling scale modifications, distortion, rotation, and other factors. |
| Keywords: KCF; target tracking; scale adaptation; multi feature fusion |
