Visual Object Tracking Based On Feature Fusion And Selection

Visual object tracking technology is one of the important research directions in the field of computer vision.It has a wide range of applications in the fields of intelligent monitoring,human-computer interaction,and military navigation guidance.In recent years,research on visual object tracking technology has emerged in an endless stream.The correlation filtering algorithm has attracted a lot of attention from researchers because of its high tracking accuracy and high speed.How to deeply mine the features makes the target tracking algorithm based on correlation filtering adapt to the requirements of various scenarios is the focus of current research.Visual object tracking is divided into single target tracking and multi-target tracking.Based on the correlation filtering,this paper improves the feature extraction module in single target and multi-target tracking process respectively:(1)In the single-objective feature extraction stage,based on the deficiencies of the existing correlation filtering algorithms,a HOG and color feature fusion algorithm is proposed based on the nucleation correlation filter.First,the characteristics of the adaptive analysis,followed by the fusion strategy,and finally experimental verification on the standard video set.(2)In the multi-objective feature extraction stage,aiming at the observation error that the existing feature fusion can bring in the multi-target tracking process,a scene adaptive feature selection algorithm based on hierarchical data association is proposed.The features are no longer just simple fusions,but distributed in each layer of the feature space.The features are gradually merged through the associated feedback of each layer until all the targets are distinguished.The experiment passes the two indicators of MOTA and FPS respectively.The scene adaptive feature selection and hierarchical data association scheme are evaluated to verify its superiority.(3)On the basis of the above-mentioned feature improvement,the paper introduces the scale estimation correlation filter,so that the feature fusion and feature adaptive selection algorithm can not only accurately predict the target position,but also achieve scale adaptation.This paper tests and analyzes the above improved algorithms through standard tracking video sets and objective tracking performance evaluation methods.Experiments show that the improved algorithm has the characteristics of fast tracking speed,high tracking accuracy and good robustness.