Research On Drone RGBT Object Tracking Algorithm Under Complex Scenes

With the development of technology,Drones are widely used in video tracking,forest search and rescue,military reconnaissance and other fields due to their flexibility and long-distance shooting.They replace traditional fixed cameras for target tracking in many scenarios.However,due to the occlusion,scale change,camera shake and many other situations in the captured video sequence,it is easy to cause the problem of the disappearance of the target,and the existing target tracking algorithm is not suitable for handling the target tracking problem of the drone in the complex scene,so the research of the drone Target tracking algorithms are of great significance in complex scenarios.Starting from the task of drone target tracking in complex scenes,this paper uses dual-light fusion to solve the problem of target tracking algorithm missing targets in complex drone scenes and improve the performance of target tracking algorithm.In this paper,firstly,through the research and analysis of single target tracking in the subtask of target tracking,the first dual-light target tracking data set for drones in this field is established,DroneRGBT,and a large number of existing target tracking algorithms are performed on the data set.了 assessment.An object tracking algorithm based on image fusion(Object Fusion Network)OFNet is proposed,which uses the feature of information complementarity between dual-light data to improve the performance of the target tracking algorithm.This method can solve the problem of target disappearance through image fusion strategy,and use unsupervised algorithm to achieve information fusion between modalities.And use the feature combination module to improve the reliability of the classification network and the accuracy of the model.Experimental results show that the algorithm proposed in this paper can improve the accuracy of the target tracking algorithm in complex scenes.Subsequently,for the improvement of the dual-light image fusion algorithm in target tracking,this paper designed a dual discriminator network(Dual Discriminator Network)DDNet with two discriminators,so that the image fitted by the generator conforms to both the visible light data distribution and the thermal infrared image Data distribution.Through the dual discriminator network,the fusion image preserves the thermal infrared target information and the texture information of the visible light image.After that,qualitative analysis and quantitative analysis of the fusion results were done on the public image fusion data set.Experimental results show that the dual discriminator fusion method can improve the performance of the tracking algorithm to a certain extent.