Fast trackling method for correlation filter using scale and support vector machine

Abstract

The invention discloses a fast trackling method for correlation filter using scale and support vector machine, which comprises the following steps of: firstly, constructing a support correlation filter by utilizing a cyclic sample, and regarding a tracking problem as a learning problem supporting the correlation filter; secondly, solving a learning problem supporting a related filter by using discrete Fourier transform and an iterative optimization strategy; meanwhile, converting a target scale estimation problem into displacement change under logarithmic polar coordinates by logarithmic polarcoordinate conversion, so that self-adaption of a target scale is realized; and finally, adopting a self-adaptive template updating strategy so that a template drifting problem under the shielding condition is solved. Therefore, tracking can be completed under the condition of giving consideration to both speed and precision in complex scenes such as severe shielding, field of view, deformation and disordered background, and the problem of tracking failure is avoided.

Description

technical field [0001] The invention relates to a video tracking method and belongs to the technical field based on machine vision. Background technique [0002] Tracking specific objects in automatically changing video sequences is a fundamental problem faced by many computer vision-themed research, such as visual analysis, autonomous driving, and pose estimation. The core problem of tracking is how to accurately and effectively detect and locate the target when the appearance of the target changes greatly due to illumination, deformation, out of view, occlusion, interference and background clutter. [0003] At present, the discriminative tracking algorithm is the most widely used tracking algorithm in the field of target tracking. It regards the tracking problem as a classification problem, and distinguishes the target from the background by training a classifier. For the discriminative classifier algorithm, one of the better effects is the correlation filter (CFs) algori...

AI Technical Summary

Problems solved by technology

Due to the relatively complex model structure and the slightly insufficient computational efficiency of the algorithm, most SVM-based trackers cannot run in real time

[0005] The above algorithms have achieved good results in practical applications, but in complex scenes such as severe occlusion, out of view, deformation, and background clutter, it is difficult for them to complete tracking with both speed and accuracy, and it is prone to tracking failures.

Images