With the rapid development in the field of artificial intelligence,image processing technology has become more and more mature.Target tracking is widely used in intelligent transportation,public place security and medical image processing.Although there have been some breakthroughs in the research of current target tracking algorithms,how to design a target tracking model that is both robust and real-time is still a challenging task for some obstructions such as partial occlusion,illumination changes and fast motion.In order to adapt to the appearance change of target tracking,this paper aims at improving the overall performance of target tracking template on the basis of inverse sparse representation framework.The main research contents include the following points:Firstly,an inverse sparse visual tracking algorithm based on local cosine similarity training weight is proposed to solve the problem of appearance change caused by occlusion and local deformation in target tracking.By calculating the local cosine similarity difference between positive and negative samples and template,Given the weight of discerning ability,selective template updating is carried out based on the weighting information comprehensively.This model can still accurately track the target in the complex environment such as partial occlusion.Secondly,in order to improve the overall performance of the target tracking model,the local structure coarse-to-fine inverse sparse target tracking algorithm based on subsampling and l2 norm is proposed by referring to the core idea of particle filter in parameter space.Through two successive template selections,Accuracy In the meantime,l2 normalized least squares is used to solve the sparse optimization problem,which increases the real-time performance of the algorithm based on the accurate positioning.Finally,aiming at the problem that the single modal feature cannot track the target very well in complex scenes,a visual target tracking algorithm based on multi-modal feature joint anti-sparse appearance model is proposed.The LBP feature and the color feature are combined with the gray feature Templates and candidate targets,which effectively solves the defects of single target representation and improves the accuracy of target tracking. |