Semi-offline deep target tracking method based on deep learning

Abstract

According to the semi-offline deep target tracking method based on deep learning provided by the invention, when the semi-offline deep target tracking method is used for target tracking, the problem of target tracking failure caused by false samples in online updating can be avoided, and the target tracking recognition performance can be improved. The method comprises the following steps: S1, constructing a tracker network model, and constructing a structure of a shared layer of the tracker network model based on an MDNet, wherein the specific domain layer is included and comprises a Dropout layer, a full connection layer and a classification function which are connected in sequence, and the feature map output by the last full connection layer in the sharing layer is input into a specificdomain layer after being processed by an activation function; s2, selecting a training set to obtain a trained tracker network model; and S3, obtaining a to-be-tracked feature sequence from the video,inputting the to-be-tracked feature sequence into the trained tracker network model, carrying out subsequent target tracking operation, training the specific domain layer through a sample collected on a first frame in the to-be-tracked feature sequence before the target tracking operation is started, and defining the network weight of the specific domain layer.

Description

technical field [0001] The invention relates to the technical field of target recognition, in particular to a semi-offline deep target tracking method based on deep learning. Background technique [0002] The target tracking technology in computer vision technology has a wide range of applications in many fields such as intelligent monitoring, human-computer interaction, unmanned driving, virtual reality and even military affairs. Based on the deep learning target tracking technology, the tracker is divided into online tracking and offline tracking according to whether to update the weight of the network model. For online trackers, the weight of the network model is more, in order to adapt to the change of the target in the tracking process, it is often necessary to adjust the weight of the network online. Among them, the MDNet (MultiDomain Network) tracker is a typical online tracker , the core idea is: first learn the general representation of the target in different type...

AI Technical Summary

Problems solved by technology

[0003] In order to solve the problem that the deep learning-based online class tracker fails target recognition due to false samples in online updates, and frequent online updates lead to target recognition performance degradation, the present invention provides a semi-offline deep target tracking based on deep learning The method, when used for target recognition, can avoid the problem of target recognition failure caused by false samples in online updates, and can improve the performance of target tracking and recognition

Images