针对在跟踪目标受到严重遮挡或者尺度变化显著等复杂场景下原始核相关滤波算法(Kernel Correlati-on Filter, KCF)追踪目标失败的问题,本文提出了一种将边缘检测算子(Edge Detection Operator, EDO)、多特征融合、尺度自适应相结合的改进型KCF目标追踪算法。首先,通过引进高斯拉普拉斯算子(Laplacian of Gaussian, LOG),对初始帧图像进行处理以获取更多边缘信息。其次,将颜色特征(Color Naming, CN)与方向梯度直方图(Histogram of Gradient, HOG)进行线性融合,可以在处理目标被遮挡时获取更多目标图像的多特征信息。然后,通过使用尺度池自适应方法解决跟踪目标时尺度变化问题。最后,使用OTB-100数据集进行算法仿真和效果评估,证明了使用本文提出的改进型KCF目标追踪算法在复杂背景下依然具有较好的准确性和鲁棒性。In this paper, an enhanced KCF target tracking algorithm is proposed to address the limitations of the original Kernel Correlation Filter (KCF) in tracking targets under complex scenarios with significant scale changes or severe occlusion. The proposed algorithm integrates Edge Detection Operator (EDO), multi-feature fusion, and scale adaptation. Firstly, Laplacian of Gaussian (LOG) is introduced to process the initial frame image for obtaining more edge information. Secondly, a linear fusion of Color Naming (CN) and Histogram of Gradient (HOG) enhances multi-feature information extraction when the target is obstructed. Furthermore, the scale change issue is addressed using a scale pool adaptive method. Finally, simulation and evaluation on OTB-100 dataset demonstrate that the improved KCF target tracking algorithm maintains high accuracy and robustness in complex backgrounds.
