Video satellite target tracking method based on background knowledge enhancement

Abstract

The invention relates to the field of video satellite target tracking, and discloses a video satellite target tracking method based on background knowledge enhancement, and the method comprises the steps: intercepting a video satellite target in a first frame image, and obtaining a target template image and a target search region image; constructing a trained double-branch Siamese network, and generating a target template feature map and a target search area feature map; constructing three cascaded RPN networks, and obtaining a screened high-confidence prediction box; and establishing a background identification module, inputting the high-confidence prediction box into the background identification module to obtain a target tracking result of the current frame image, training the background identification module on line, and updating the scene category of the background identification module. Background enhancement is carried out through background offline training, background trainingset updating, background online updating and the like, the method is suitable for different complex backgrounds, background interference is removed to the maximum extent, the problem that a target issubmerged in the background is solved, and the target is tracked more accurately.

Description

technical field [0001] The invention relates to the field of video satellite target tracking, in particular to a video satellite target tracking method based on background knowledge enhancement. Background technique [0002] Video satellites shoot ground objects from a high altitude, and their imaging methods, observation angles, and spatial resolutions are significantly different from those of ground videos, resulting in fewer pixels for moving objects in satellite videos, lack of texture information, and lack of features of moving objects. Compared with terrestrial video data, video satellite images have lower contrast, and the distinguishability and identifiability of moving objects and backgrounds are weaker. These differences lead to the fact that moving target detection and tracking algorithms for terrestrial video cannot be well adapted to satellite video. [0003] Classic target tracking algorithms include: tracking algorithms based on recursive Bayesian, kernel den...

AI Technical Summary

Problems solved by technology

Although such algorithms have achieved good tracking performance on typical target tracking databases, they usually require manual design of features, which affects the generalization ability of the algorithm to a certain extent.

In addition, in the video satellite target tracking task, the main difficulty is background interference and occlusion.

The existing video satellite target tracking algorithm cannot perform background enhancement, and the existing target tracking algorithm cannot be applied to different complex backgrounds, cannot remove background interference, and cannot solve the problem that the target is "submerged" in the background, and the accuracy of target tracking is low

Images