Research On Related Technologies Of Ground Target Tracking Based On Airborne Vision

Ground target tracking technology based on airborne vision not only plays an important role in modern warfare,but also plays an irreplaceable role in forest fire prevention and urban security,and the key is mobile target tracking algorithm.With the increase of social demand and the progress of science and technology,people's demand for UAV tracking ground targets is higher and higher.This also puts forward higher requirements for tracking ability of the algorithm.From the characteristics of airborne vision,based on lots of experiments,a variety of mobile target tracking algorithm for precise analysis of the validation,selection can be applied to the airborne vision ground target tracking,and the tracking effect is relatively good target tracking algorithm for the further research.After a lot of experimental analysis,aiming at the shortcomings of the selected target tracking algorithm,we made targeted improvements to make it better meet the requirements of target tracking based on airborne vision.The main contents of this paper are as follows:1.The environmental influencing factors of drone tracking ground targets were analyzed,and the image preprocessing algorithm was selected.Drones in the air will be affected by natural factors such as the wind,resulting in severe jitter,which will cause video jitter;In the transmission of video signals,the received image will usually contain Gaussian noise;these influencing factors,usually Affects the effect of target tracking,so image preprocessing is required to obtain a higher quality output image.2.Aiming at the shortcomings of DSST(Discriminative Scale Space Tracking)target tracking algorithm,which is suitable for airborne vision and has good tracking characteristics and is sensitive to the target deformation,the color features and the(Histogram of Oriented Gradient)HOG feature are combined to improve the tracking ability.The DSST algorithm has a high dependence on the spatial distribution of the target pixel,and is very sensitive to the target deformation.Therefore,when the target undergoes severe deformation,the tracking effect is poor.The color histogram is only a statistical probability distribution of the color,so it is not highly dependent on the spatial distribution of the target pixel and is not sensitive to the deformation of the target.Therefore,in order to solve the problem of tracking failure caused by drastic deformation of the target when the UAV is tracking ground non-rigid targets,the original DSST algorithm is improved,and the color features and the HOG characteristics are effectively combined.3.An improved model updating strategy is designed for DSST algorithm when the target is obstructed and the tracking fails.During the tracking of ground targets,drones are often obstructed by trees,lamp posts,etc.If the model is still updated at this time,too many error messages will be introduced,which will lead to tracking failure.Therefore,the model is not updated when the target is severely obstructed,and the model is updated when the model update condition is satisfied.4.For the shortcomings of the improved algorithm's slow running speed,the running speed of the algorithm is improved by improving the target search strategy and adopting feature dimensionality reduction methods.When the drone tracks the ground target,it needs the algorithm to have a good running speed,but the above improvement introduces some calculations and reduces the running speed.Therefore,In order to improve the speed of the algorithm,a new search strategy was designed,when the tracking situation is good,a local search is performed,and when the situation is not good,a global search is performed.At the same time,through data dimension reduction,feature dimensions are reduced,calculations are reduced,and algorithm operation speed and real-time performance are improved.