Target tracking method and system based on attention mechanism and global reasoning

Abstract

The invention discloses a target tracking method and system based on an attention mechanism and global reasoning, and belongs to the technical field of computer vision, and the method comprises the steps of carrying out the target tracking through a target tracking model based on a twin network, wherein the target tracking model comprises a template branch and a search branch, the template branch and the search branch each comprise a backbone network, a parallel attention mechanism and a global reasoning module; obtaining an initial frame picture and a current frame picture, taking the initial frame picture and the current frame picture as input of the template branch and input of the search branch respectively, and obtaining a first score graph and a second score graph; performing weighted summation on the first score graph and the second score graph to obtain a regression graph; and determining the position of a target according to the regression graph. Compared with an existing tracking algorithm, the method has a better tracking effect.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to an object tracking method and system based on attention mechanism and global reasoning. Background technique [0002] Object tracking is one of the hard problems in the field of computer vision, which is the basis for higher-level visual understanding and scene analysis. Object tracking technology is widely used in video surveillance, human-computer interaction, robotics, video editing, and autonomous driving. The task of visual object tracking is to achieve continuous and stable tracking of moving targets in subsequent frames based on the target position and size information in the initial frame. Achieving long-term stable object tracking is still a challenging task due to object disturbances such as scale variation, rotation, deformation, fast motion, and changes in background illumination of the object. [0003] In recent years, research on the task of visual trackin...

AI Technical Summary

Problems solved by technology

For the typical twin networks SiamFC and SiamRPN, the following points out two shortcomings. First, the network structure they use is shallow, the extracted features are not sufficient, and they do not pay attention to the tracking target itself, so in the face of some tracking There will be a tracking failure during the challenge

Second, none of them takes context information into account, and it is easy to cause tracking failure in the face of objects with large occlusion or excessive deformation.

Images