Twin neural network moving target tracking method based on full-connection attention module

Abstract

The invention discloses a twin neural network moving target tracking method based on a full-connection attention module, and belongs to the technical field of computer vision tracking. According to the method, after the image features are extracted by using a twin neural network, template features extracted by the template branches are processed by using a full-connection attention module, and the template features are fused with the original template features and then serve as the attention-enhanced template features to be combined with updated template features for performing the same operation; the obtained new template features are fused with search features, so that self-attention and mutual attention of the template features are realized, and the robustness is improved; position information and size offset information of a target in the corresponding search graph are obtained according to a response graph fusing the reinforced template features and the search features; and the input of the update template branch is updated according to a network prediction result of each fixed frame, so that the tracking precision is improved. According to the method, target tracking can still be continuously and stably realized under the conditions of severe deformation, reproduction after transient disappearance or shielding and the like of the target.

Description

technical field [0001] The invention relates to a method for tracking a moving target in an image sequence, belonging to the technical field of computer vision tracking. Background technique [0002] Moving object tracking technology is one of the important research directions in computer vision science, and has a wide range of applications in video surveillance, human-computer interaction, intelligent navigation and other fields. This technique refers to the ability to predict the location of objects in subsequent frames given the bounding box of the object in the first frame of a video sequence. At present, the main problems of moving target tracking technology are the influence of complex interference factors such as illumination change, target occlusion, shape change, size change and fast movement, which make it difficult to achieve real-time tracking and robust and accurate target tracking methods. [0003] In recent years, deep learning has achieved great success in t...

AI Technical Summary

Problems solved by technology

But at the same time, there are still problems in the target tracking algorithm of the Siamese neural network framework. At present, the relatively mature algorithms such as SiamFC, SiamRPN, and SiamBAN only obtain the target template from the first frame. The lower tracker is relatively less robust

Images