Visual tracking method and visual tracking system under moving background

Abstract

The invention discloses a visual tracking method and a visual tracking system under a moving background. The visual tracking method comprises the following steps: reading a first frame of a video and labelling a target and a tracking area; circularly and densely acquiring the labelled target and an image of the tracking area, extracting a feature; taking a context area Xi around the target as a basic training sample, adding upper and lower frame constraints, and obtaining a learning classifier increment coefficient a; circularly moving a detection area z in the next frame to obtain a candidate block, substituting into a classifier to achieve that f(z)=(K<z>)<T>a and a target position that (xi, yi)=arg max f(z); returning to the step B, taking a context where the target center (xi, yi) is located as the basic training sample in the step C, obtaining the classifier increment coefficient a, updating a model, and continuing to update the next frame. The visual tracking system comprises a target initializing module, an image feature extracting module, a classifier initial training module, a rapid target detecting module, and a classifier training module. The tracked target can be accurately tracked when the target rapidly moves and the scale changes.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a visual tracking method and a visual tracking system with low calculation intensity, wide application range, and the ability to accurately track target fast-moving and scale-changing motion backgrounds. Background technique [0002] Machine vision is an extremely important part of modern manufacturing, involving interdisciplinary subjects in many fields such as artificial intelligence, neurobiology, psychophysics, computer science, image processing, and pattern recognition. In computer vision, how to accurately know the position or even the size, direction, and shape of the target in the video image is a basic problem, and it is also a problem to be solved by video tracking technology. The purpose of target tracking technology is to locate information such as the position, state, trajectory, scale, and speed of the target in a given video or image sequence, a...

AI Technical Summary

Problems solved by technology

However, the classifier learned by the kernel correlation filter tracking algorithm does not consider the spatio-temporal relationship between two consecutive frames. When the target surface changes suddenly, the learned discriminative classifier is difficult to robustly represent the change of the target surface, and it is easy to cause tracking drift.

In addition, since the kernel correlation filter tracking algorithm does not consider the scale problem, the scale change of the distance between the camera and the target position during the movement will also cause the follow-up tracking to shift during the target tracking process.

Images