A Kernel Correlation Filtering Target Tracking Method Based on Superpixel and Hybrid Hashing

Abstract

The invention provides a kernel correlation filtering target tracking method based on ultra-pixels and hybrid Hash. The method is characterized in that a target appearance model is reconstructed through an ultra-pixel clustering method and an ultra-pixel segmentation method, a target is divided into meaningful ultra-pixel blocks, and the ultra-pixel block parameters of each ultra-pixel block are calculated, and the effective features of each ultra-pixel block are extracted; Gaussian kernel correlation filtering-based tracking operation is performed, so that the tracking results of candidate ultra-pixel blocks are obtained; the LAB color Hash sequence and DCT Hash sequence of each ultra-pixel block are calculated and are combined into a hybrid Hash sequence; and based on the ultra-pixel block parameters and geometric constraint correction positions, the location of the target to be tracked is positioned, and the scale of the target to be tracked is estimated. With the method adopted, the tracking of the whole target can be realized, the accuracy and anti-jamming ability of the tracking of the target can be improved, and the scale change problem of the target can be effectively solved.

Description

technical field [0001] The invention relates to the technical field of image processing and analysis, more specifically, to a method for tracking a target with kernel correlation filtering based on superpixels and hybrid hashing. Background technique [0002] As one of the most active topics in the field of computer vision, object tracking plays an extremely important role in video surveillance, human-computer interaction, behavior analysis, military operations and other fields. In recent years, object tracking has been greatly developed, and researchers in the field of computer vision have proposed a large number of object tracking algorithms. The current target tracking methods can be roughly divided into two categories: one is to generalize the target tracking problem into a binary classification problem, and to distinguish the target from the background by training a binary classifier to achieve continuous tracking of the target; The first is to learn the appearance mod...

AI Technical Summary

Problems solved by technology

Although this method consumes less time and is faster, its robustness still needs to be improved.

[0005] It can be seen that there are still many deficiencies in the existing algorithms in the field of target tracking, especially the mutual constraint relationship between target tracking accuracy and speed still needs to be comprehensively solved

Images