AR (Augmented Reality) target tracking method for improving KCF (Kernel Correlation Filtered Wave) algorithm

Abstract

The embodiment of the invention provides an AR (Augmented Reality) target tracking method for improving a kernel correlation filter (KCF) algorithm. The method comprises the following steps: preprocessing a video image; performing feature fusion in combination with the gradient feature HOG and the texture feature LBP; tracking a target area based on a KCF algorithm of improved feature fusion; performing image matching on the window target and the current video frame by using an rBRIEF algorithm; and calculating a three-dimensional registration matrix and superposing virtual information to complete augmented reality three-dimensional registration. The problem that the tracking effect is affected due to unstable tracking and easy loss can be avoided.

Description

technical field [0001] The present application relates to the technical field of image recognition, in particular to an AR target tracking method that improves the Kernel Correlation Filtering (KCF) algorithm. Background technique [0002] AR (Augmented Reality, augmented reality) technology [1] It is an emerging research field developed on the basis of VirtualReality technology. It has the characteristics of virtual reality combination and real-time interaction. It is an enhancement of real scenes rather than a replacement. [2] , the key to building an augmented reality system is to accurately superimpose virtual information on the target area of ​​the real world, that is, three-dimensional tracking and registration technology [3] . [0003] 3D tracking and registration technology uses graphic image recognition to accurately fuse and register virtual information and real scenes. KCF (Kernelized Correlation Filter, kernel correlation filter) algorithm [4] , originating f...

AI Technical Summary

Problems solved by technology

Using the KCF algorithm, because the target frame has been set during the tracking process, the size changes from the beginning to the end, but the target size changes during the tracking sequence, which will cause the target frame to drift during the tracker tracking process. , which leads to tracking failure; secondly, KCF does not solve the processing problem when the target is occluded during the tracking process.

The area to be registered will have problems with local occlusion and changes in lighting conditions, has no rotation invariance and scale invariance, and is sensitive to noise

Therefore, the registration accuracy is low, which affects the matching effect

Images