Convolutional neural network regression model-based visual tracking method and device

Abstract

The present invention belongs to the computer vision field and proposes a convolutional neural network regression model-based visual tracking method and device. The invention aims to solve the problem that a target tracking process is divided into two independent steps of component matching and target positioning and cannot infer the location of a target directly through components. The method comprises the following steps that: S1, an image block is sampled according to a given target to be tracked at the initial frame of visual tracking, the image block is divided into a plurality of components; S2, a pre-constructed convolutional neural network regression model is trained through using a random gradient descent method; and S3, at the follow-up frames of visual tracking, a search area is constructed based on the position of the target to be tracked in the last frame, and the position of the target to be tracked in a current frame is obtained through the trained convolutional neural network regression model. According to the method and device of the invention, the components are fully combined with the positioning of the target, and therefore, better robustness can be realized.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a visual tracking method and device based on a convolutional neural network regression model. Background technique [0002] Visual tracking is the most fundamental component in computer vision applications, and fields such as intelligent video surveillance, augmented reality, robotics, and human-computer interaction all require robust tracking of objects of interest. Although great progress has been made in this field in recent years, visual tracking is still a difficult task because it faces challenges such as partial occlusion, deformation, lighting changes, motion blur, fast motion, background clutter, and scale changes. Wait. [0003] In order to solve these problems, many component-based methods have been extensively studied in recent years, which decompose the target object into a set of components for study. In fact, some parts of the target remain visible when...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the above-mentioned problems in the prior art, that is, in order to solve the problem that the target tracking process is divided into two independent steps of component matching and target positioning, and the position of the target cannot be directly inferred from the components, one aspect of the present invention proposes A visual tracking method based on a convolutional neural network regression model, comprising the following steps:

Images