Research On Object Tracking Algorithm Based On Correlation Filter

Tracking,as one of technologies in computer vision,has been widely used in various fields.In many kinds of tracking algorithms,tracking algorithms based on correlation filters have attracted much attention due to their rapidity and superior tracking ability..Therefore,this thesis is devoted to the research of tracking algorithms based on correlation filters.Firstly,this thesis analyzes and summarizes many object tracking algorithms based on Correlation Filter in recent years.Secondly,this thesis analyzes ECO(Efficient Convolution Operators)in detail.Because ECO cannot track the object again when the object is completely occluded,an improved ECO algorithm based on response confidence template update strategy is proposed in thesis.Two kinds of response confidence evaluations are designed to automatically adjust the learning rate of template.In addition,experiments were carried out on two public datasets.The experiments show that the improved algorithm can still track the object when the object is completely occluded.And compared with the original algorithm,it has better tracking success rate and accuracy rate in terms of object scale change,object blur and object deformation.Finally,in order to overcome the shortcomings of ECO in object deformation,this thesis proposes an improved ECO algorithm with likelihood map,which uses Bayesian pixel classification model to obtain likelihood map of the object,and then fuses the likelihood map with other features.Similarly,experiments were carried out on public dataset.The experiments show that the improved algorithm has better success rate and accuracy rate in terms of object deformation,object low resolution and object rotation.