Visual Tracking By Using Local Feature And Hyper-graph Matching

A new method for visual object tracking is proposed in this paper?Visual tracking algorithm can be divided into four modules: feature extraction,target representation scheme,search mechanism and model update.Tracking algorithms must have the ability to handle the complete disappearance of the object for an undefined amount of time.The target is represented by a three-layer structure in this paper.The target is represented by a point set of ORB feature in the first layer.In the second layer,the target is divided into several local regions,each local region is represented by an HSV color histogram.The position of each local region is determined by the particle filter.Local features often have less information than global features,but the relationship between the local features can help track the object.The paper use high-order graph matching method to local feature tracking to optimize the tracking results of local features.In the third layer,a naive Bayesian classifier is maintained.The classifier describes the global color characteristics of the target.By using spatio-temporal Context,the influence introduced by backgrounds can be eliminated.Three layers work together to deal with appearance change caused by abrupt motion,illumination variation,shape deformation,and occlusion.The object model is updated every frame to account for appearance changes.In order to verify the performance of the proposed tracking algorithm,this paper does a lot of experiments on the public test dataset and compare with some excellent tracking algorithm published in recent years.Experimental results show the effectiveness of the proposed algorithm.