A Correlation Filter Tracking Algorithm Incorporating Time Information

Abstract

A target tracking algorithm that incorporates time information into the relevant filter is proposed, LTCF. LTCF is a combination of LT model and CFM model. We first analyze from the tag data that the target has time continuous characteristics and relative time information during the movement. Then, the LT model is designed on the basis of LSTM algorithm, which mainly to predict the time information about the target of the next frame. Finally, the information which predicted form LT model is combined with the CFM model, which improves the calculation time of the filter on the one hand, and reduces the overall algorithm parameters on the other. Experiment results show that compared with STRCF and SRCF, the accuracy of the LTCF algorithm is improved by 7% and 11% respectively, and LTCF algorithm model also reduces the model size while ensuring real-time tracking.