Target tracking method based on twin neural network and parallel attention module

Abstract

The invention discloses a target tracking method based on a twin neural network and a parallel attention module, and belongs to the field of machine vision. The method comprises the following steps: 1, cutting a template image and a search area image according to the position and size of a target in a video sequence picture to form a training data set; 2, constructing a twin network, wherein a basic skeleton of the twin network adopts a finely tuned residual network; 3, embedding parallel attention modules into template branches of the twin network, wherein the parallel attention modules comprise a channel attention module and a space attention module which are parallel to each other; 4, based on the training set, constructing an adaptive focus loss function, training a twin network with aparallel attention module, and obtaining a network model with training convergence; 5, performing online tracking by using the trained network model. In the tracking process, the problems of target appearance change and the like can be effectively solved, and the tracking precision is improved.

Description

technical field [0001] The invention belongs to the field of machine vision, in particular to a target tracking method based on a twin neural network and a parallel attention module. Background technique [0002] With the extensive research of machine vision in theory and practice, object tracking has gradually become a basic but crucial branch. The task of target tracking is to calculate the specific position of the target in each subsequent frame only based on the bounding box of the target in the first frame, so various objective factors such as object deformation, occlusion, fast motion, blur, lighting changes, etc. making tracking a challenge. At present, target tracking can be mainly divided into methods based on correlation filtering and methods based on deep learning. For a long time before deep learning became popular, most target tracking algorithms were based on correlation filtering. Although this type of algorithm greatly reduces the computational cost through...

AI Technical Summary

Problems solved by technology

The dual-branch structure of the Siamese network is relatively ingenious, but there are still several problems to be improved: (1) In the feature extraction part of the original Siamese network, the feature expression ability of the shallow neural network is weak, and the advantages of deep learning are not fully utilized. ; (2) The loss function used in the training process is easily affected by simple samples

Images