Research On Target Tracking Algorithm Based On Global Motion Trajectory

The importance of single target tracking research in the field of computer vision is becoming increasingly prominent,and it is widely used in real life,such as intelligent transportation,intelligent monitoring,and unmanned driving.Current target tracking research is mainly based on detection-based tracking and tracking based on twin network comparison.These algorithms can better deal with challenges such as background interference,lighting changes,target deformation,etc.,but they are not very effective for target occlusion.In view of this situation,this paper proposes the idea of combining “detection-based tracking” and “global target trajectory assisted tracking” to meet the challenge of target occlusion.The specific work is as follows:(1).Aiming at the problem of target trajectory acquisition,this paper proposes a global target trajectory acquisition algorithm based on image stitching.Specifically,the tracking video sequence is stitched into a “panoramic image” based on the SIFT image mosaic method,and the motion trajectory of the target in the “panoramic image” is sequentially obtained.Then use the least squares method to fit the relative coordinates of the target to obtain the corresponding motion trajectory,and then predict the next movement trend of the target.(2).For trajectory assisted tracking problems,this article actually assists tracking by improving the confidence of the target candidate samples in the area Where the movement trend is located.Specifically,two weighting factors Q and W are proposed.The former is used to increase the weight of target candidate samples that conform to the trend of the movement;the latter is used to reduce the confidence interference generated by the original algorithm on the one hand,and to avoid the The probability problem caused by Q.(3).In particular,aiming at the objective problem of the lack of target occlusion challenging video sequences in the current tracking research data set,it is also to verify the correctness of the ideas mentioned above.In this paper,we propose a new OFV-25 dataset with target occlusion as a challenge,which contains 25 video sequences with a total of 3900 frames.In the experimental part,this paper tests the proposed algorithm with other classic algorithms of the same period on the OTB-100 dataset.The experimental results verify the correctness of the idea of assisted tracking based on the global target motion trajectory.The test on OFV-25 dataset highlights the effectiveness of thealgorithm and dataset.