Research On Object Tracking Algorithms Based On Adversarial Transfer Learning

As one of the research hotspots in the field of computer vision,object tracking technology has a wide range of applications in video surveillance,intelligent driving,and human-computer interaction.Although experts and scholars at home and abroad have proposed a variety of algorithms,due to the complexity and variability of application scenarios of object tracking technology,facing challenges such as lighting changes and object occlusion,the tracking results are not satisfactory,Therefore,efficient object tracking technology still needs urgent research Based on the perspective of adversarial transfer learning and siamese networks,this thesis improves the existing object tracking methods from the aspects of sample enhancement and feature selection.The main work is as follows:(1)This paper proposes an object tracking algorithm based on adversarial transfer learning.Aiming at the problem of insufficient training samples in the object tracking field,a transfer learning method is introduced to transfer the target features of the source domain to the target domain,so that the algorithm can learn the general features of object tracking.The use of generative adversarial networks to enhance data samples solves the problem of high overlap of positive samples during online training of the algorithm and difficulty in capturing large-scale deformation.The algorithm is verified on the OTB2013 dataset and the OTB2015 dataset.From the experimental results,it can be seen that the object tracking algorithm proposed in this paper has better performance than some existing advanced object tracking algorithms.(2)This paper proposes an object tracking algorithm based on the attention mechanism.Aiming at the problem that Siam FC network based on template matching strategy leads to insufficient feature discrimination ability,we introduce channel attention mechanism and spatial attention mechanism,so that the algorithm pays more attention to the features that are favorable for object tracking in the spatial location and channel location.An online update strategy is added to fuse the image features of the first frame with the image features with higher confidence in subsequent tracking image frames,reducing the probability of the tracking failure when the target encounters challenges such as occlusion and deformation.Experimental results on the OTB2013 dataset and OTB2015 dataset show that the proposed algorithm has higher accuracy.