Research On Unmanned Aerial Vehicle Target Tracking Algorithm

In recent years,deep learning has developed rapidly,and researchers have begun to use convolutional neural networks for target tracking.In addition,with the rapid development of unmanned aerial vehicle technology,the existing unmanned aerial vehicle products have the advantages of simple operation and low cost,and the use of unmanned aerial vehicles for video target tracking has become a new research hotspot.For example,in the field of public safety monitoring,compared with fixed camera monitoring,UAVs are flexible and can obtain richer target information,making it easier to continuously lock and track targets.However,due to the rapid changes in the flying attitude,flight speed,and flying height of the UAV,problems such as the relative movement of the camera and the tracking target,the background information interfering with the target,and the small tracking target will occur during the information collection process.Therefore,it is very difficult and challenging to apply target tracking technology to videos taken by drones.In order to build a UAV video target tracking algorithm with higher tracking accuracy and success rate,this paper proposes a UAV video target tracking algorithm based on multi-domain adversarial learning and a UAV video target tracking algorithm based on multi-domain attention regularization..The main work of this paper is as follows:Aiming at the problem that insufficient positive samples and strong discriminative features of a single frame in UAV video target tracking can easily lead to overfitting of the classifier,a UAV video target tracking algorithm based on multi-domain confrontation learning is proposed.Introduce the generative adversarial network into the feature generation of multi-domain learning,and use adversarial learning to improve the robustness of feature extraction;add hole convolutions with different expansion coefficients to the convolutional layer for multi-scale feature extraction,and build with different receptive field feature extraction module;add modulation factor to the cross-entropy loss function to solve the problem of imbalance in the number of positive and negative samples.Experimental results show that the tracking accuracy and success rate of the algorithm are improved.Aiming at the problem that the tracking algorithm in UAV video target tracking is difficult to adapt to the significant movement of the target and its robustness in the time dimension,a UAV video target tracking algorithm based on multi-domain attention regularization is proposed.Use the partial derivative of the network output with respect to the input image as the attention map,take the attention map as the regular term,and combine the original loss function to train the network.The experimental results show that the tracking network that introduces attention regularization can adapt to the fast movement of the target,and the tracking effect is excellent.