Target Real-time Tracking Algorithm Based On Improved SIFT

As the accuracy and real-time requirements of the target tracking algorithm continue to increase in the application scenario,the related target tracking algorithm also needs to be continuously improved.The application scenario of this thesis is to use the uploaded company logo or the employee's business card as the target image in the augmented reality APP,track the target image in real time in the video and overlay the company information above the target image.In this thesis,the situation that the target object may be partially occluded and the target scale changes during the tracking process is deeply studied.(1)An improved SIFT(Scale Invariant Feature Transform)target detection algorithm is proposed to determine the initial position and tracking window of the target object and optimize the algorithm.Firstly,the rules for screening feature points are improved according to the specific scenarios of the algorithm.The algorithm reduces the computational complexity and maintains the accuracy of the algorithm.Secondly,the FREAK descriptor is used instead of the SIFT descriptor to reduce the vector dimension.Finally,A clustering feature tree algorithm is proposed to improve the feature matching process and reduce the computation time.(2)In this thesis,the pyramid LK optical flow method is used to calculate the optical flow vector by using the SIFT feature points of the target obj ect.This method reduces the calculation time and improves the accuracy.(3)In this thesis,the rules for dynamically updating SIFT feature point sets and feature point weights are proposed,which improves the robustness of the algorithm when the target part is occluded.The target real-time tracking algorithm based on improved SIFT proposed in this thesis achieves the goal of reducing computation time and improving accuracy.When the target object has partial occlusion or scale change,the algorithm has good accuracy and robustness,and can also achieve the goal of real-time tracking.