Research On Robust Correlation Filter Tracking

Visual object tracking is a fundamental and important research direction in computer vision,which is widely applied in industrial community.Although visual object tracking has made great progress in recent years,it still faces many problems to be solved.In this paper,we conduct a research on robust visual object tracking based on the correlation filter,and the main work is as following.We propose a visual object tracking algorithm,which is combined with Pixel-based Color Model and Correlation Filter(PCMCF).The Pixel-based Color Model(PCM)is a collaborative tracking module,which can make up for the shortcomings of the correlation filter in the complex senses such as occlusion and deformation.The reliability of the color feature affects the performance of the PCM module.Two feature description methods are proposed to describe the pixel color features,which are based on fixed interval quantization and adaptive interval quantization separately.In addition,the online learning strategy plays an important role in visual object tracking.We propose an adaptive online learning strategy based on the score map of the Pixel Color Classifier,which makes the tracker more robust.We propose a Soft Mask Correlation Filter(SMCF)to deal with the boundary effects in the correlation filter.The soft mask makes the circulant shifting conducted on a larger image patch,and crops a mass of real background patches for training.The soft mask is constructed by spatial prior probability and appearance likelihood probability.The soft mask enables the correlation filter to pay more attention to the center part of the target.In addition,we propose an adaptive updating strategy based on High-confidence Criterion.The reliability of the correlation filter response map can be evaluated by the confidence score,and the update strategy is adaptively adjusted,which improves the robustness of the tracker.Quantitative and qualitative evaluations on challenging benchmark sequences demonstrate that the proposed PCMCF and SMCF tracking algorithm perform favorably against most state-of-the-art algorithms.Our tracking algorithms achieve more robust performance in the complex senses of occlusion,deformation,out-of-view and background clutters.