Research On Target Tracking Algorithm Based On Visual Attention Mechanism

As a classic problem in computer vision,the target tracking algorithm has attracted many domestic and foreign research scholars in the past decades.The goal tracking algorithm has also made breakthrough progress and has been widely used in life and work.This technology has developed over decades,and has successively improved and developed classic tracking algorithms such as Kalman Filtering,Particle Filter,and Mean Shift algorithm.Among them,Mean Shift tracking algorithm is the most classic algorithm in single target tracking algorithm.Mean Shift tracking algorithm has been widely concerned by scholars because of its small computational amount and insensitive to edge occlusion,rotation,deformation and other advantages.However,when background interference or sudden changes occur during the tracking process,the disadvantages of Mean Shift tracking algorithm are also obvious.This paper proposes an improved Mean Shift tracking algorithm based on saliency features and ORB matching algorithms for background interference.At the same time,it proposes an improved Mean Shift tracking algorithm based on template updates and linear prediction for problems that cannot be tracked in real time under sudden changes.Under the premise of the above disadvantages,real-time accurate tracking of the tracking target is achieved.The main research contents of this paper are divided into the following three aspects:1.Preprocess the image,and explain the theory,mathematical model,tracking process,advantages and disadvantages of Mean Shift tracking algorithm.The image preprocessing consists of three parts: image smoothing and denoising,color space transformation and color space selection.2.Aiming at the problem that traditional Mean Shift tracking algorithm cannot accurately track in real time under complex background,an improved Mean Shift tracking algorithm based on saliency features and ORB matching algorithm is proposed.First,the frequency-tuned MSSS saliency area detection algorithm is improved.The improved algorithm can extract more significant saliency maps.The saliency maps are combined with the back-projection maps extracted by Mean Shift tracking algorithm for subsequent tracking.Then,aiming at the problem of target loss that may occur in the tracking algorithm under similar background conditions,the ORB feature point matching algorithm is used to form a target finding mechanism.The Bhattacharyya coefficient is used to determine whether the tracking target is lost,and then to determine whether to enable the target finding mechanism.By comparing and analyzing the tracking effect and performance of the test video,theimproved algorithm is more robust against background interference and can suppress similar background interference.3.Aiming at the problem that the traditional Mean Shift tracking algorithm cannot accurately track in real time under sudden changes,Mean Shift tracking algorithm based on template updating and linear prediction is proposed based on visual attention mechanism.This article first introduces a template update mechanism.This mechanism introduces a background template based on the original target template.The target template and background template are compared with the set threshold to determine the interference factors.When the template update is needed,select the appropriate speed to update the template.Then,when the interference factor determines that the target is blocked,a linear prediction equation is introduced to predict the target position information,which effectively solves the tracking problem in the case of target blocking.By comparing and analyzing the tracking effect and performance of the test video,the improved algorithm has better anti-interference ability than the traditional algorithm under sudden changes.