Target tracking method based on correlation filtering fusion convolution residual learning

Abstract

The invention relates to a target tracking method based on correlation filtering fusion convolution residual learning, and belongs to the technical field of computer vision tracking. The target tracking method comprises the following steps: firstly, extracting a training block taking a target object as a center from an input frame for model training; during tracking, extracting a search block fromthe predicted central position of the previous frame, and positioning a target by searching the maximum response value; updating the width and the height of the target object at the T frame accordingto the generated response mapping range; using training blocks and response diagrams collected for each frame as training pairs to be input into the network to be updated on line by utilizing residual learning, so that the prediction performance of the model in target motion is effectively maintained. According to the method, the correlation filter is expressed as a convolution layer, and featureextraction, response generation and model updating are integrated into the neural network for end-to-end training, so that semantic information of the to-be-tracked target is utilized to the greatestextent, and the tracking precision is improved.

Description

technical field [0001] The invention belongs to the technical field of computer vision tracking, in particular to a target tracking method based on correlation filter fusion convolution residual learning. Background technique [0002] Visual target tracking, commonly known as single target tracking, first gives a human-labeled rectangular frame in the first frame, and then requires a tracking algorithm to follow the frame in subsequent frames. It belongs to an important research direction in computer vision and has a wide range of applications, such as: video surveillance, human-computer interaction, unmanned driving, etc. The human eye can easily follow a specific target for a period of time. But for the machine, this task is not simple, especially in the tracking process, there will be various complex situations such as the target undergoes severe deformation, is blocked by other targets, or similar objects interfere. The deep model has demonstrated its powerful capabili...

AI Technical Summary

Problems solved by technology

However, the high-level features mainly reflect the semantic characteristics of the target, and are more robust to the apparent changes of the target, and the precise positioning of the target sometimes drifts

And large-scale online fine-tuning slows down the tracking speed and affects the frame rate

Images