Research On Planar Object Tracking Algorithm Based On Depth Feature Description And Optical Flow

Plane object tracking is a hot research topic in computer vision field and it is widely used in tasks such as augmented reality,robot navigation and video surveillance.In recent years,many algorithms have been used to solve the problem of planar object tracking,but most algorithms only consider the correspondence between the template image and the target image for the input video sequence,ignoring the motion information of the object during the tracking process.This limits the improvement of the robustness of the algorithm to a certain extent.In view of this,this paper applies the motion information(optical flow information)of the target object to the planar object tracking task,and is committed to exploring more accurate planar object tracking algorithms.The research work in this paper is as follows:(1)In order to obtain the motion information of objects between two adjacent frames in the video,an optical flow estimation algorithm based on dense Siamese network is designed.The traditional optical flow algorithm estimates the optical flow information according to the change of the image pixel value,which generally needs to be based on the assumption of constant brightness and small motion,which greatly limits the application scope of the algorithm.In this paper,the densely connected convolutional neural network is used as the feature extraction network and the feature map provided by it is used as the input,the optical flow inference network is designed and trained to provide the optical flow information between two adjacent video frames.(2)A planar object tracking algorithm based on deep feature description and optical flow is designed and implemented.The algorithm uses the optical flow information between two adjacent frames to adjust the position of the corresponding points of the template image key points on the target image,thereby improving the accuracy of the tracking algorithm.In the implementation stage of the algorithm,this paper integrates the existing feature extraction network,optical flow inference network and Point-of-interest Matching Layer into a unified network architecture to provide the optical flow information and corresponding point matching functions required by the tracking algorithm.Since the same feature extraction network is shared,the feature map does not need to be repeatedly calculated during the tracking process,which reduces the memory usage to a certain extent and improves the tracking speed of the algorithm.(3)The proposed algorithm is verified by experiments.In this paper,the proposed algorithm is implemented in Python language,and evaluated on POT dataset,UCSB dataset and TMT dataset.Finally,the tracking results of the algorithm on POT dataset are visualized.The experimental results fully demonstrate the effectiveness of the proposed planar object tracking algorithm based on deep feature description and optical flow.At the same time,it also shows that the strategy of using optical flow information to improve the planar object tracking algorithm has a certain feasibility.