Anti-shake UAV Visual Guidance System Based On Dual-modal Target Trackin

UAV visual guidance system is an important research content of intelligent application at present.The existing rigid connected UAV system is difficult to recognize,lock and approach ground targets in complex scenes,and it is more difficult to deal with imaging jitter,rotation,ground similar interference targets and other problems.This paper designs and builds an anti shake UAV visual guidance system based on dual-mode target tracking.The system is composed of UAV,camera,embedded core board and other hardware.It is equipped with software modules such as target detection,tracking and positioning to realize target recognition,locking and approach in complex scenes.Aiming at the problem that the existing target tracking module is difficult to meet the needs of UAV visual guidance system in complex scenes,this paper focuses on the optimization of the target tracking module in the system software flow.A dual-mode tracking algorithm(siamdl)based on fusion attention of feature level and decision level is proposed and studied.This bimodal tracking algorithm is based on the siam network structure design,and the prediction frame is regressed with anchor-free.Firstly,a multi-domain awareness module is proposed by introducing the cross domain siam attention mechanism.This module interacts with the context information of different time domains and modal domains,so as to fine tune the template features.Then,a dual-level balance module is proposed by introducing the dual-level fusion attention mechanism.This module uses featurelevel information and decision-level information to balance the bimodal fusion features.The siamdl algorithm is trained by taking the classification constraints of single modal features and the classification and regression constraints of fused modal features as the loss function.Compared with the benchmark algorithm,the tracker can deal with low illumination and hot cross scenes well.The tracker maintains high speed,and the accuracy and success rate on Las He R data set are improved by 4.7% and 3.5% respectively.An anti-shake UAV visual guidance system based on siamdl is built.According to the requirements of mission scenario,this paper makes a targeted design of software and hardware,and determines that the hardware framework is composed of image input module,embedded core board module and UAV flight module.The image input module is responsible for collecting images,the embedded core board module is responsible for carrying the software framework,and the UAV flight module is responsible for flying and carrying other hardware.According to the task scenario and selection,the software framework is composed of image acquisition module,serial port transceiver module,target detection module,target tracking module,coupling decision module and positioning module.In the software framework,this paper focuses on the target tracking module responsible for dual-mode and anti-shake tracking,this module includes dual-mode tracking algorithm(siamdl)and anti-shake tracking algorithm(transtracker),in which the anti-shake tracking algorithm is realized through feature perspective transformation and positioning correction operation.The above software and hardware framework are combined to form an anti-shake UAV visual guidance system based on siamdl.Finally,this system is tested in continuous locking,approaching target scene and jitter,rotation and similar interference target scene.This system can deal with complex scenarios stably and efficiently,and the proportion of successful tests has reached 93.3%.