Target tracking method and system based on multi-feature and self-adaptive dictionary learning

Abstract

The invention discloses a target tracking method and system based on multi-feature and adaptive dictionary learning, and the method comprises the steps: obtaining a foreground template of a first frame image at an initial position of a target and a surrounding background template of the target; Constructing an initial dictionary according to the target foreground template and a target surroundingbackground template, and obtaining an initial texture feature dictionary and an initial color feature dictionary; according to the LC-KSVD algorithm, Performing learning processing on the initial texture feature dictionary and the initial color feature dictionary by using a KSVD algorithm to obtain a discriminant texture feature dictionary and a discriminant color feature dictionary; And performing target position tracking matching based on a sparse feature matching algorithm according to the discriminant texture feature dictionary and the discriminant color feature dictionary to obtain a tracking position of a next frame of the target. In the embodiment of the invention, by utilizing the discrimination dictionary, the representation capability of the target appearance can be supplemented,background noise interference is reduced, and the tracking precision is improved.

Description

technical field [0001] The invention relates to the technical field of target tracking, in particular to a target tracking method and system based on multi-feature and adaptive dictionary learning. Background technique [0002] In the field of computer vision, the visual tracking problem is still a challenging study due to its complex scenes, such as object occlusion, object deformation, rotation, scale changes, and cluttered backgrounds. In recent years, there have been many improved visual tracking algorithms based on different theoretical frameworks. Most of these algorithms can be roughly divided into generative models and discriminative models. The generative model usually finds the most similar candidate or finds the candidate with the highest similarity to the original target in subsequent frames, but when the object is matched or the prediction algorithm is not accurate enough, This tracking algorithm is easily overruled. The discriminative model classifies the for...

AI Technical Summary

Problems solved by technology

Most of these algorithms can be roughly divided into generative and discriminative models. Generative models usually find the most similar candidate or the candidate with the highest similarity to the original object in subsequent frames, but when the object matches or the prediction algorithm is not accurate enough, This tracking algorithm is easily overruled

Images