Research On Long-term Object Tracking Method Based On Multiple Correlation Filtering Model

Object tracking technology is widely used in various fields of production and life.With the wider application of object tracking technology,how to achieve robust tracking of object has become a core issue in the field of object tracking.Since the object tracking algorithm based on correlation filtering was proposed,it has been favored for its high tracking accuracy,high success rate and fast calculation speed.However,most correlation filtering algorithms cannot track the object continuously for a long time,and cannot realize the adaptive update of the model.Therefore,this thesis conducts an indepth study on existing tracking framework,focuses on the object relocation module and update mechanism of the model,and aims to better achieve the tracking of long-term object.The main research work is as follows:(1)In the tracking process,inspired by the long-term memory and short-term memory of the human brain,on the basis of the existing multiple correlation filtering model algorithm framework,a long-term filter for confidence detection of tracking targets and an online detector for correcting target positioning are added in the tracking process.The algorithm framework of long-term correlation filter,short-term correlation filter,and online detector are designed together to complete the confidence detection of the object position and the modification of the object position.(2)Aiming at the problem of fixed update of the model in the algorithm,this thesis adopts the peak sidelobe ratio to provide a certain degree of confidence in the tracking results based on the above algorithm framework,that is,the peak sidelobe ratio is used to quantify the accuracy of object positioning.The learning rate is adjusted through the average value of the frame difference between two adjacent frames,the object displacement between two adjacent frames,and feature changes.Then,by setting the corresponding learning rate in segments to achieve adaptive model update to make sure the model can discern the object in different situations.(3)The algorithm in this thesis is compared with various algorithms on the authoritative data sets OTB-2013 and OTB-2015.The experimental results show that the algorithm in this thesis effectively improves the stability of tracking,and obtains a highquality long-term tracking effect in a complex situation with a variety of challenge factors.