Unmanned aerial vehicle video target tracking method based on deep learning

Abstract

An improved unmanned aerial vehicle target tracking algorithm comprises four steps of S1, S2, S3 and S4. A new deep neural network structure is firstly designed, and a deep neural network structure capable of well representing object features is obtained through offline learning. Then, online tracking is carried out, firstly, a motion model between two adjacent frames is obtained through feature points, then target features are extracted through the deep neural network structure obtained through offline learning, a deep neural network feature with good target edge information description capability and certain degree of robustness to noise and geometric deformation is provided by using a focus loss function, a generative tracking method is designed by performing motion compensation on adjacent frames, and the method can obtain a good tracking result under various scenes such as camera rapid motion, video jitter, light ray change and scale change, and is suitable for various civil and military systems such as face tracking, unmanned aerial vehicle tracking and military target tracking systems.

Description

technical field [0001] The present invention relates to the technical field of computer vision and image processing, in particular, to a deep learning-based UAV video target tracking method. Background technique [0002] Traditional UAV tracking methods use color histograms as features (I.F.Mondragon, P.Campoy, M.A.Olivares-Mendez, and C.Martinez, 3D object following based on visual information for Unmanned Aerial Vehicles, in IX Latin American Robotics Symposium and IEEE Colombian Conference on Automatic Control, pp.1-7, 2011.), the effect of target tracking depends on the description of the target by color features, so for similar color target tracking, the effect is not very good (C.Teuliere, L.Eck, and E. Marchand, Chasing a moving target from a flying UAV, in 2011 IEEE / RSJ International Conference on Intelligent Robots and Systems, IEEE, September, pp.4929-4934, 2011.). Morphological filtering technology is also a method of UAV tracking (A.Wainwright and J.Ford.Fusion ...

AI Technical Summary

Problems solved by technology

[0003] The UAV-based target tracking mentioned above has three shortcomings. First, the number of videos given in the experiment is small and the video time is short, so it is difficult to effectively judge the pros and cons of the tracking algorithm.

Secondly, the features used by the algorithm are traditional features, which cannot model the target well, so the tracking effect is not good

Third, because the UAV has a large monitoring range, and because the consumer UAV is small and unstable, when shooting video, the UAV platform is quite unstable, so the video is prone to jitter. under, it is easy to lose track of the target

Images