Research On UAV Target Tracking Algorithm And Application Based On Camshift

Computer vision has always been a hot research field,and video target tracking is one of the very important research directions.Both in the military field and the civilian field have high research significance and a wide range of application scenarios,such as: imaging guidance,Intelligent video surveillance,intelligent transportation,intelligent medical treatment,human-computer interaction,etc.Therefore,video target tracking has attracted more and more scholars to study it and produced many excellent algorithms.Among them,the target tracking algorithm based on CamShift is a continuous adaptive mean shift algorithm,which is based on color histogram feature tracking Compared with MeanShift algorithm,this algorithm can well solve the problem of object size change during tracking,so it has become a research hotspot for video target tracking.At the same time,with the maturity of UAV technology,UAV target tracking has also become A hot topic.Therefore,in this paper,the application of using drones in the nature protection work of Qilian Mountains is studied.Due to the high altitude shooting and the particularity of mountain terrain,it is particularly important to design a robust tracking algorithm.When using the CamShift algorithm for tracking,three serious problems will be encountered.One is that the algorithm simply considers the color histogram,but the drone will change the lighting when shooting the target,which will seriously affect the tracking effect.The second is due to the relative movement between the drone and the tracked target,and the tracking angle is always changing,which will affect the tracking accuracy.The third is that the aerial photography is at an overhead angle,and the target is easily blocked,which will also cause Poor target tracking.In order to select a target tracking algorithm that is well adapted to the drone tracking situation and improve the tracking accuracy,this article proposes an effective solution,mainly including the following research content:(1)When the drone is tracking the target,both the lens and the target are susceptible to light,so the HLBP(Haar local binary pattern,HLBP)feature that is not sensitive to light changes is used as the target matching feature.When the lens is backlit or the target enters the shadow area,the HLBP feature is used to improve the tracking robustness.Experimental data shows that the introduction of HLBP feature can solve the effect of light changes on tracking and improve the tracking accuracy of the algorithm.Aiming at the problem that the UAV cannot always track the target fromone angle during aerial photography,that is,the rotation angle of the target changes,this paper will use HLBP's rotation invariance to improve the tracking accuracy.Experimental data shows that the introduction of HLBP feature can solve the problem of target rotation,At the same time,for the problem of fixed threshold in the process of HLBP extraction,this paper proposes an adaptive method of HLBP threshold to perform adaptive threshold tracking for different situations.(2)When the UAV is tracking the target,due to the complex terrain,the target is prone to occlusion,which leads to the problem of tracking failure.In this paper,the Kalman filter algorithm is introduced to predict the position of the target,and the occlusion judgment is based on the Pap distance.Experimental results show that the introduction of kalman prediction algorithm can improve the algorithm's anti-occlusion ability.(3)Apply the improved algorithm to the video taken by the drone on the Qilian Mountain protection area,and verify the algorithm through the real scene.