Correlation filtering tracking algorithm based on fusion features and self-adaptive updating strategy

Abstract

The invention relates to a correlation filtering tracking algorithm based on fusion features and a self-adaptive updating strategy. A tracking method based on correlation filtering is adopted, and a robust target model is constructed by using sub-channel fusion features to predict the central position of a target. For the problem that a target is severely shielded, a self-adaptive model updating mechanism is introduced to solve the problem that firstly, the reliability degree of a current response graph is judged according to a confidence coefficient threshold value, and the higher the reliability degree is, the lower the possibility that the target in the image is shielded is. Constructing a self-adaptive updating function on the basis, updating the tracking model according to the function, ensuring that the model is updated at a very low learning rate when the target is severely shielded, and introducing noise as little as possible. When the appearance of the target is clear and interference factors such as shielding deformation do not exist, the model is updated at a very high learning rate, and it is guaranteed that the model can capture the latest target characteristics. By means of the measures, a very robust tracking result can be obtained under different challenging scenes.

Description

technical field [0001] The invention relates to a target tracking method, which belongs to the field of computer vision. Background technique [0002] At present, aerial video tracking technology has been widely used in military and civil fields. Compared with videos shot by fixed platforms or handheld devices, aerial videos have their own unique properties. First of all, the camera moves at high speed with the UAV, and there are transformations such as translation and rotation between the images of the aerial video sequence, and the scenes in the video are complex and changeable, and the target is easily disturbed by occlusion and noise; in addition, because the UAV sometimes Flying at a high altitude of thousands of meters or even tens of thousands of meters, the proportion of moving objects in the image is very small, which brings great challenges to aerial video processing. In recent years, a large number of tracking methods based on correlation filtering have emerged,...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problem that the appearance model drifts due to the severe occlusion of the moving target in the aerial video, which easily leads to tracking failure, a robust and efficient target tracking method is designed.

Images