Research On Airport Moving Object Tracking Based On Multi-feature Fusion

In recent years,with the rapid development of air transportation industry,the operating pressure of airport surface activity area is increasing.Traditional surveillance methods can not meet the requirements of safe and efficient surface surveillance gradually.With the development of computer vision technology,video-based surface surveillance technology has gradually emerged as an effective auxiliary surveillance method,and object tracking technology is the key link.In the actual airport surface environment,object tracking often faces the influence of factors such as occlusion,rotation,scale variation,low resolution,background clutter,etc.,resulting in reduced tracking accuracy or even loss of tracking object.Aiming at the above problems,this dissertation analyzes the image features and object tracking algorithms,and proposes an airport object tracking algorithm based on multi-feature fusion.The specific work of this dissertation is as follows:First of all,the theories and applications of traditional target tracking algorithms,correlation filter based target tracking algorithms and deep learning based target tracking algorithms are summarized respectively,and the practicability of each algorithm in the surface surveillance is analyzed.Then,the principles and characteristics of various image features are analyzed.An interpolation operator is introduced to fuse the manual features and deep features of the target to enhance the ability of feature expression of target in low-resolution environments and with small-size.A simulation experiment of the tracking algorithm using single feature and fusion feature is carried out respectively on the airport video data set,and the influence of each feature on the tracking result is compared and analyzed.Finally,aiming at the lack of tracking accuracy determination mechanism and the low efficiency in filter updating of the original algorithm,a tracking result confidence evaluation mechanism is proposed,and on this basis,an adaptive filter model update strategy is designed so that the filter can update when the result is reliable,and the learning rate can be adjusted adaptively.The simulation results of the airport video data set show that compared with the baseline algorithm,the overall precision of the algorithm is increased by 11.05%,the success rate is increased by 10.84%,and the overall performance is better than other algorithms.