Research On Single Object Tracking Algorithm Based On Computer Vision

Computer Vision is a new research field that explores how to get information from images(or videos)for analysis and understand,the object tracking based on vision is an important and challenging research direction in the field of computer vision.The goal of visual object tracking is to detect,recognize and track the moving targets in sequences,and obtains the relevant parameters of targets,such as the location,velocity,scale and moving trajectory,a behavior understanding or other higher-level tasks can be realized by farther analysis.Visual object tracking is the basis to solve other computer vision tasks.With the ever-increasing demands of video analysis in modern society,the object tracking has attracted more and more researchers,and also made great progress in recent years.But in practical tracking process,due to the uncertainty of moving target and the complexity of application scenarios,it is still a challenging task to propose an object tracking algorithm with good real-time performance,accuracy and robustness.To solve the above problems,with deep analysis of the framework and principle of the traditional object tracking algorithms,we propose some innovative and valuabletracking methods according to the generative and discriminant observation model.Those methods can effectively improve the accuracy of tracking results and efficiency of tracking process,which satisfies the current actual demands.The main research work and achievements in this paper are summarized as follows:1)Based on the generative observation model and sparse representation theory,we present a robust tracking method based on the inverse sparse group lasso model(ISGL).The visual tracking models based on sparse representation usually use template set as dictionary atoms to reconstruct candidate samples.This conventional sparse representation method needs to solve multiple objective function optimization problems without considering the similarity among atoms.In order to reduce the process of solving optimization problems and enhance the tracking performance,the templates are encoded by the candidate samples in this paper,and similar samples selected to reconstruct the template at group level,which facilitates inter-group sparsity.The optimal tracking result is decided by the group sparsity coefficients of candidate sample.Our inverse sparse group lasso model combines the advantages of inverse sparse representation and group sparse representation,decreasing the running time,reducing the reconstruction error and improving the tracking performance.In order to enhance the performance of our algorithm,an observation model based on multi-information...