Research On TLD Target Tracking Algorithm Based On Spatio-temporal Context

During the target tracking process,when the target moves quickly or is severely blocked,the tracking frame will have a certain offset from the real target,even if the target is lost.The main purpose of this paper is to improve the accuracy of the target tracking by combining the advantages of the Tracking-Learning-Detection(TLD)algorithm,the Spatio-Temporal Context algorithm(STC)and the Kalman filte.It can accomplish in tracking the target when it moves quickly and is occluded.The specific work as follows:1.This article briefly introduces the STC and the TLD algorithm.It explains the operation flow of the STC algorithm in detail.It can help us increase the understanding of each model in the algorithm by studying the application of the Fourier transform in the STC algorithm.This article describes the operation process of the TLD algorithm in detail,it describes the specific principles of the detector tracker and learner,and analyzes the performance of the TLD algorithm.2.There is a phenomenon that the target tracking is lost when the target moves fast or is occluded in the STC tracking process.Based on the STC algorithm,a robust STC-KF algorithm is designed.STC-KF uses Kalman filtering to estimate the target.The followings are the procedure:Firstly,it uses the STC algorithm to track the target.Next,it determines whether the target is lost according to the distance of two consecutive frames.If the target is lost,Kalman filter estimates the target location.This article compares this chapter algorithm and STC algorithm through a series of experiments.3.In order to solve the continuous tracking problem in the target tracking process and improve the tracking ability of the algorithm,based on the TLD target tracking algorithm framework,this paper designs TLD-STC-KF algorithm who has better tracking effect.The core of the algorithm is to preserve the advantages of the TLD algorithm and use the STC-KF algorithm replaces the median flow tracking algorithm in the TLD algorithm.If the target is lost.Kalman filter estimates the target location.After the target reappears,the learner learns the target characteristics,updates the detector parameters,improves the tracker performance,and continues to track the target,it implements algorithmic continuous target tracking.Finally,this article compares this chapter algorithm and STC-KF,TLD algorithm through a series of experiments.