Twin network target tracking method based on deformable convolution

Abstract

The invention discloses a twinning network target tracking method based on deformable convolution, and the method specifically comprises the steps: constructing a twinning backbone network model: employing the deformable convolution to carry out the learning of the image features of a template branch and a search region, so as to obtain corresponding local semantic information, aggregating the local semantic information to obtain global context related information; and constructing a feature fusion network model: calculating the similarity between the template branch features and the search branch features through cross-correlation operation, and performing target tracking on the target candidate block with the maximum similarity score. According to the method, the global matching accuracy of the target image and the target image of the search area is improved, and more accurate tracking is realized.

Description

technical field [0001] The invention relates to a twin network target tracking method based on deformable convolution, and belongs to the technical field of target tracking. Background technique [0002] Object tracking has always been a basic research topic of visual computer, and it is widely used in the fields of video surveillance, intelligent driving and intelligent medical diagnosis. Target tracking mainly includes target tracking algorithms based on correlation filtering and target tracking algorithms based on deep learning. The target tracking algorithm based on correlation filtering uses the video itself as the training data, which limits the expressive ability of the model. Although the speed is fast, the tracking accuracy is difficult to improve. The tracker based on the twin neural network can better solve the problem of correlation filtering accuracy, and the speed of the related algorithm has also been improved, which has received extensive attention from rese...

AI Technical Summary

Problems solved by technology

[0004] In view of the problems existing in the above-mentioned prior art, the present invention provides a twin network target tracking method based on deformable convolution (A twin network target tracking method based on deformable convolution, DeSiamFC), which utilizes deformable convolution to make full use of context information, Learn sample offset, and modulate by learning feature amplitude to extract target features more accurately to obtain a more robust tracking model, which can solve the problem of target deformation and occlusion

Images