Object Tracking Method Based on Self-Attention Transformation Network

Abstract

The invention discloses a target tracking method based on a self-attention transformation network, which is realized by using a search image encoding module, a target encoding module, a decoding module, and a supervision module; the search image encoding module is composed of a multi-head self-attention network and a feedforward The network is implemented in series, and the search image block encoding is calculated through the multi-head self-attention network and the feed-forward network; the target encoding module calculates the target feature encoding through the masked multi-head self-attention network; the decoding module is based on the target feature encoding, through the multi-head attention network in the search In the image feature encoding, the query matching is performed to calculate the position coordinates of the target prediction frame; the supervision module calculates the error between the two according to the position information of the target prediction frame and the real target position information, and the neural network parameters obtained when the error is minimized are used for target tracking. The invention has a stable tracking effect, is easier to capture deformed targets in search images, and generates accurate tracking results.

Description

technical field [0001] The invention relates to the field of target tracking methods, in particular to a target tracking method based on a self-attention transformation network. Background technique [0002] With the development and popularization of video equipment, many fields based on analyzing video have emerged, such as security monitoring, traffic control, autonomous driving, etc. The development of these fields will greatly improve the quality of life. Object tracking technology is one of the important components in the field of video analysis. Although it has achieved excellent tracking results, it often fails to track in the face of complex actual scenes. Therefore, the current object tracking technology needs to further improve its ability to adapt to the actual scene. [0003] In the target tracking task, the target to be tracked is arbitrarily specified, and the shape of the target is also arbitrary. In order to cover objects of arbitrary shapes, current tracki...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problem of a priori anchor frame in the existing video object tracking method or avoid the use of a priori anchor frame, the effect of the video object tracking method in the test set or practical application is not sensitive to the parameter setting of the a priori anchor frame , so as to have a stable and accurate tracking effect. The present invention discloses a target tracking method based on a self-attention transformation network, which is realized by using a search image coding module, a target coding module, a decoding module, and a supervision module;

Images