Correlation filtering tracking algorithm based on significance detection and robustness scale estimation

Abstract

The invention relates to a correlation filtering tracking algorithm based on significance detection and robustness scale estimation. The method comprises steps that filter training, and a cycle matrix training filter is constructed through utilizing the target prior information; significance detection, l2 regularized constraints of an original regularized ridge regression problem are converted into l2 and 1 constraints; secondly, a significance area is extracted from a candidate area through utilizing boolean map saliency (BMS); target position prediction, relevant filtering is carried out for a sample containing the significance information, the maximum value of a filter response map corresponds to the position of a target, and parameters of the filter are updated; robustness scale estimation, two-dimensional scale change of the target in length and width can be acquired through a two-dimensional scale filter. The method is advantaged in that significance detection is introduced under the relevant filter framework, scale change of the target can be acquired through employing the two-dimensional scale filter, and tracking accuracy and the success rate are improved.

Description

technical field [0001] The invention belongs to the field of image processing, in particular to target tracking based on correlation filtering. Background technique [0002] As an important research direction in the field of computer vision, target tracking refers to the analysis of a series of continuous video images acquired by visual sensors such as cameras to obtain the position, size, and motion status of a specific target or multiple targets. Useful information. A basic vision tracking, the goal is to use the initialization labeled object box to predict the trajectory after the object in the image sequence. In the tracking process, due to some external factors, such as illumination changes, shooting angle changes, target occlusion, etc., the main challenges in the current tracking are: the deformation of the target due to its own motion or external factors, and being blocked by other objects during motion. , changes in scene illumination and motion blur caused by tar...

AI Technical Summary

Problems solved by technology

The occlusion problem has always been a difficult problem in tracking. Its causes are complex and diverse, and there is no prior information. In addition, it is difficult to identify the occlusion itself

Therefore, it is necessary to estimate the changes of the length and width of the target separately, but no one has realized it based on the correlation filter.

Images