Multi-scale filtering target tracking method based on adaptive feature fusion

Abstract

The invention provides a multi-scale filtering target tracking method based on adaptive feature fusion, and relates to the technical field of target tracking. The method comprises the following steps: initializing a network model, and reading a video to obtain a plurality of image frames; target position prediction: firstly, selecting image blocks around a target, extracting VGG-19 network layer features and CN features, broadcasting the CN features, then directly connecting and fusing the CN features with conv2-2 layer features of the VGG-19 network in series, classifying the three layers of features through a classifier, outputting a response through a filtering model, and finally evaluating the target position by using empirical weight weighting fusion response from coarse to fine, and updating the target filter parameters. According to the method, filtering is carried out by combining the hierarchical features of the VGG19 network with the color attribute CN features, responses are output respectively, the maximum response is taken as the reference, and the target position is evaluated from coarse to fine weighted fusion responses. The OPE detection result shows that the method is reliable and high in precision in the aspects of target deformation, illumination variation, motion blur and plane rotation.

Description

technical field [0001] The invention relates to target tracking in the field of artificial intelligence, in particular to a multi-scale filtering target tracking method based on adaptive feature fusion. Background technique [0002] With the rapid development of artificial intelligence technology, the application of object tracking technology in real life in video surveillance, human-computer interaction, motion analysis, activity recognition, etc. is becoming more and more extensive. At the same time, it also faces problems such as rapid target movement, target occlusion, complex background changes, shape changes, and noise interference. [0003] Common target tracking algorithms are mainly divided into two categories, one is the target tracking algorithm based on depth features, the other is the target tracking algorithm based on correlation filtering, and the target tracking algorithm based on depth features, deep features can extract good images features, but the traini...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to solve the problems of KCF target tracking algorithm in the tracking of target deformation, rotation, motion blur, etc., and improve the accuracy and success rate of KCF tracking algorithm in target tracking

Images