Research On A Fusion Strategy Of Visual Attention Feature For Object Tracking

With the rapid development of computer vision,visual object tracking has increasingly attracted the attention of academia.Visual object tracking also plays a vital role in applications such as automatic driving,intelligent transportation,intelligent security,and human-computer interaction.However,visual object tracking still face enormous challenges.On one hand,it is the current visual tracking algorithm that is difficult to handle the challenges in the actual scenes such as object occlusion and illumination.On the other hand,most visual tracking algorithm design ideas originate from the object detection algorithm,which ignored time-space relativity of motion object in continuous time.In addition,how to ensure the balance between speed and accuracy of the algorithm is also the focus of research.Therefore,how to deal with these challenges has become an important research topic in computer vision.Inspired by the human visual attention mechanism,this thesis proposes a strategy of fusion visual attention feature with object tracking.This strategy attempts to solve the problem that visual tracking algorithm cannot use time-space relativity,and at the same time deals with the challenges of the actual scenes.Finally,the effectiveness of the strategy was verified in benchmark datasets.The major contributions of the thesis as follows:1)Explain the motivation of the visual attention feature,and put forward the concept of visual attention feature and its composition.On this basis,the differences and connections between visual attention features and hand-craft features are analyzed.Then explain the impact of visual attention features on the tracking algorithm.In addition,a simple and efficient visual attention feature extraction method is introduced.2)Analyze the advantages and disadvantages of discriminant tracking algorithms,and propose a generalized fusion strategy for visual attention features.Based on this strategy,the typical representative of correlation filter and Tracking-by-Detection algorithms,KCF and PF+HoG,are selected to fuse the motion charecteristics and saliency of visual attention feature.3)On visual tracking algorithm evaluation benchmark OTB2013 and OTB2015,two improved algorithms for visual attention features are evaluated.Finally,qualitative analysis and quantitative analysis are used to verify the effectiveness of the proposed object tracking strategy for fusion visual attention features.At the same time,it proves that the fusion visual attention feature can raise the baseline algorithm to state-of-the-art.