A Visual Tracking Method Based On Siamese Network And Ridge Regression

Visual tracking is not only a basic research content in the field of computer vision,but also plays a key role in the application fields of automatic driving,human-computer interaction,intelligent robot systems and smart city systems.With the in-depth study of deep learning,visual tracking has also achieved great success,and has achieved amazing results in multiple public test sets.Although the offline model based on deep learning has achieved success in tracking accuracy and speed.When the offline model encounters a target that has not appeared in the training data set during the tracking process,it easily ends in failure.The introduction of an online model that is updated in real time to obtain the status information of the target during the tracking phase is an important way to make up for the shortcomings of the offline model.How to make the online model and offline model complement each other's advantages is essential to improve the accuracy and success rate of the tracking algorithm.First,introduce a local matching mechanism based on the SiamFC model to make up for the lack of global matching of the SiamFC model,and complete the offline model training in an end-to-end manner.Secondly,this paper embeds the closed-form solution of the ridge regression method into the deep model to provide good initialization for model optimization and obtain an efficient online tracking model.Finally,the efficient response graph fusion mechanism combines the two to obtain a tracker with better performance,which is called the JONF model.Experimental results show that the JONF model proposed in this paper can greatly improve the accuracy and success rate of the tracking model.