Research On Single Target Tracking Methods In Satellite Videos Equipment

With the development of satellite remote sensing technology,the detection and tracking of targets using video satellites plays an important role in intelligent traffic monitoring,military security applications and public safety monitoring.Compared with the target tracking of traditional video,there are problems in satellite video such as lower image resolution,inconspicuous feature information,small target size and low contrast with the background,etc.Traditional target tracking algorithms cannot be well applied to satellite video,and it is important to study the target tracking algorithms applicable to satellite video,and specific work includes the following aspects:(1)Based on the kernel correlation filtering algorithm,we combine the motion information,spatio-temporal information and appearance model of the target to achieve the segmentation and tracking of a single target.Firstly,the target position is predicted using the trained kernel correlation filter to obtain target candidate region one;then the spatio-temporal model of the target and its surrounding area is built based on the colour space features to calculate the place with the highest likelihood probability as target candidate region two;then the visual background extraction algorithm is used to detect the target region on a pixel-by-pixel basis to obtain target candidate region three;these three candidate regions are evaluated separately for The three candidate regions are evaluated separately,and the best region is used as the final prediction location;finally,the model discriminant condition is used to adaptively calculate the target global model update speed to achieve more robust tracking.(2)Based on the SiamCAR algorithm,the single target tracking is achieved by integrating the attention mechanism and the motion information of the target.A motion excitation module and a channel attention module are first added to the feature extraction network to enhance the target’s feature information;then adjacent frames are added as new templates to the algorithm to form a triple network to supplement the template information;then the target trajectory is predicted using the Kalman filter algorithm,and the predicted templates are added to the algorithm to form a quadruple network to increase the target’s motion information.Tracking targets in satellite video under the constructed quadruple network model makes full use of the complementary information to obtain the optimal position of the target,effectively improving the accuracy of target tracking.(3)Design of a visualisation software to visualise target tracking results,segmentation results,real-time video frame rates and target position coordinates by correlating improved kernel correlation filtering algorithms and visual background extraction algorithms to provide users with better data analysis and decision support.The improved algorithm is compared with the benchmark algorithm on a dataset in the satellite video target tracking domain.The experimental results show that the tracking performance of the optimised algorithm achieves significant improvements in accuracy and success rates compared to the benchmark algorithm.