Target tracking method and device based on the all -convolutional neural network

Abstract

The embodiment of the invention discloses a target tracking method and device based on a fully convolutional neural network. In a pre-training phase, object nature discrimination and object verification are carried out after the fully convolutional neural network is utilized to extract image features of training sample images, and a fully convolutional neural network model is obtained through training; then parameters contained by the fully convolutional neural network model are utilized to build an online tracking network; and finally, the online tracking network is utilized to track a targetin a video on which tracking is to be carried out. According to the method, an object nature discrimination process is added in the pre-training phase, an algorithm is enabled to learn what is an object, and the problem that the algorithm is not robust for a noisy background is solved; at the same time, an object verification process is added, that is, two objects are given for judging whether the two objects are the same object, therefore, object verification more emphasizes differences between the objects, and weakens differences between classes, thus the target is separated from a background and all possible interfering objects, and finally, an accuracy rate of target tracking is increased.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to a target tracking method and device based on a fully convolutional neural network. Background technique [0002] Object tracking is an important part of video analysis technology, that is, given the bounding box of an object in the first frame of a video, the tracking algorithm needs to automatically find this object in subsequent videos. [0003] The tracking target may have large-scale deformation, lighting changes, interfering objects, and occlusions throughout the video, and the input accepted by the tracking algorithm is only the bounding box of the user in the first frame, which requires the tracking algorithm to have self-learning ability and be able to distinguish Interfering objects and occlusions. In recent years, with the rapid development of deep learning algorithms, such algorithms (for example, convolutional neural networks) have also been introduced into...

AI Technical Summary

Problems solved by technology

[0005] An embodiment of the present invention provides a method and device for target tracking based on a fully convolutional neural network to solve the problem of low target tracking accuracy in the prior art

Images