Learning target tracking algorithm based on twin network

Abstract

The invention discloses a learning target tracking algorithm based on a twin network. The method comprises the following steps: 1, setting a function f to compare the feature map similarity of a template image x and a candidate image y, the maximum response value corresponding to a target position, the x being generally an image with a target as the center in a first frame, and the y being an image searched by taking an (n-1) th frame of image target as the center in an nth frame; substituting the two inputs into a convolutional neural network, learning a network parameter rho by using a convolution processing function, generating two cross-correlation mappings, and performing exhaustive search on x at y to respond to the position of a target corresponding to the maximum value; wherein animproved discriminant correlation filter layer is added between input x and cross-correlation operation in the twin network, 3, performing tracking by measuring the similarity between a template imagex and a search image y, and online fine adjustment is carried out by means of a discriminant correlation filter module during online tracking so as to ensure the tracking accuracy.

Description

technical field [0001] The invention relates to the field of information technology, in particular to a twin network-based learning target tracking algorithm. Background technique [0002] Target tracking is to mark the first frame of the target in a video frame, and then locate the position of the target in each frame of the video to generate the target motion trajectory. The problem of target tracking plays a key role in real-time vision applications in the fields of intelligent surveillance systems, automatic driving, UAV monitoring, intelligent traffic management, human-computer interaction, etc., and is an important content of computer vision research. [0003] Object tracking has become a research hotspot in the field of computer vision because of its wide application in real life, but it still faces many problems. With the great success of deep learning in the field of object recognition and tracking, more and more researchers have started to apply the convolutional ...

AI Technical Summary

Problems solved by technology

At present, the improvement of tracking performance based on DCF methods is mainly carried out in the aspects of multi-dimensional features, robust scale estimation, nonlinear kernel, long-term memory components and reducing boundary effects. However, these methods reduce the tracking speed while improving the accuracy.

Images