TY - EJOU AU - Huang, Detian AU - Gu, Peiting AU - Feng, Hsuan-Ming AU - Lin, Yanming AU - Zheng, Lixin TI - Robust Visual Tracking Models Designs Through Kernelized Correlation Filters T2 - Intelligent Automation \& Soft Computing PY - 2020 VL - 26 IS - 2 SN - 2326-005X AB - To tackle the problem of illumination sensitive, scale variation, and occlusion in the Kernelized Correlation Filters (KCF) tracker, an improved robust tracking algorithm based on KCF is proposed. Firstly, the color attribute was introduced to represent the target, and the dimension of target features was reduced adaptively to obtain low-dimensional and illumination-insensitive target features with the locally linear embedding approach. Secondly, an effective appearance model updating strategy is designed, and then the appearance model can be adaptively updated according to the Peak-to-Sidelobe Ratio value. Finally, the low-dimensional color features and the HOG features are utilized to determine the target state to further improve the robustness of the tracker. The experimental results on OTB-2015 benchmark validate that the proposed tracker can effectively solve the illumination variation, scale variation, partial occlusion and deformation in the complex background. KW - Tracking KW - Appearance model KW - Color attribute KW - Locally linear embedding KW - Multi-scale DO - 10.31209/2019.100000105