Research And Application Of Improved Kernelized Correlation Filter Tracking Algorithm

Visual tracking is an important research direction in the fields of computer vision and video security monitoring.In this thesis,the performance improvement of kernel-based correlation filter tracking algorithm is studied.An adaptive model updating factor and fusion tracking algorithm with sub-region are proposed to improve tracking accuracy,adaptability and robustness.To verify the performance,a simulation experiment system is built on cross-platform with MATLAB and C++,and completes the practical application test.Aiming at the problem that the update model of correlation filter tracking uses a fixed learning factor,an adaptive learning factor is defined with tracking response strength,the model update weight can be adjusted in time according to the tracking quality.Experiment results verify the effectiveness of adaptive learning factor.To solve the problem that the target tracking model is easily contaminated when the target appears occlusion,deformation and other unexpected situations,advanced sub-region fusion tracking algorithm is proposed in the thesis.The tracking target is divided into multiple sub-regions,and multiple independent trackers based on adaptive learning factor are used respectively for tracking.According to the tracking intensities of each sub-region,the fusion weight of sub-regions is constructed for position and scale estimation.A model updating strategy based on high tracking confidence is designed to improve the stability of the algorithm.Finally,the proposed tracking algorithm is verified on simulation and practical application.Experiments on the tracking dataset show that the proposed algorithm has a significant improvement on the overall accuracy and tracking robustness.The results demonstrates excellent performance on multiple situations such as occlusion,deformation,scale variation et.Furthermore,a real-time tracking application system is implemented based on the proposed algorithm.Practical tests show that the system can meet the real-time requirements and track the target accurately.