Method and device for tracking video object

Abstract

The invention discloses a method and device for tracking a video object. The method of the invention comprises steps of obtaining a training sample, wherein the training sample is a target object window of the target object in a first frame image of an input video; obtaining a label of the training sample, wherein the label of the training sample is a gauss label matrix which is constructed according to characteristics of a rotation matrix of a target object window; calculating related filtering parameters for tracking the target object according to the label of the training sample; performing detection and tracking on the target object in the next frame image according to the related filtering parameters; updating the training sample according to the tracking result of the target object in the next frame image and returning to the step of obtain the training sample. The method for tracking a video object realizes detection and control on the size changing of the target object, further improves the tracking efficiency, and has high practicability, adaptability and robustness.

Description

technical field [0001] The invention relates to the technical field of computer multimedia, in particular to a video object tracking method and device. Background technique [0002] Video object tracking technology is a basic problem in the field of computer vision, and has a wide range of applications in many real-world scenarios, such as security monitoring, human-computer interaction, and automatic control systems. Generally speaking, the video tracking problem can be simplified as an image online learning problem, so how to build an accurate, efficient and maintainable tracking model is the most important step in video tracking technology. Related video tracking methods can be divided into two categories, one is the generative model, which searches for the region most similar to the target object in the video; the other is the decision model, which uses an image classifier to distinguish the foreground object from the background region , so as to find the target object ...

AI Technical Summary

Problems solved by technology

[0004] Judging from the current research, the problem of variable-scale object tracking has become one of the difficulties faced by correlation filtering methods.

If the target object has a large scale change during the tracking process, a single correlation filter tracker cannot scale appropriately, which will eventually lead to tracking failure

However, the existing improved methods often use multiple filters or multi-layer scale space pyramids to deal with this problem. Although they can detect the scale changes of the target object to a certain extent, but because of the multiple filtering operations in a single frame which greatly reduces the tracking efficiency

Moreover, if the scale change in the tracking process is not controlled, it may cause more serious offset and false detection, and reduce the robustness of the tracker.

Images