Research on Person Object Tracking Method Based on Extended Reality

In the process of designing extended reality simulation systems for military operations based on Extended Reality, common challenges include slow processing of high-resolution images, poor real-time integration of virtual and real elements, and tracking failures during rapid movement of characters. To address these issues, we propose an improved single-object tracking and detection algorithm that combines YoloV7 and Kernelized Correlation Filters (KCF), achieving lightweight and high-precision computation. Firstly, based on the principles of the KCF algorithm, a Multi-Scale Module is designed to perform multiple upscaling and downscaling operations on input sequential frame images, obtaining feature-rich yet smaller-sized images, these images are further smoothed and denoised using Gaussian filtering. Additionally, to tackle the problem of KCF's inability to update tracking boxes promptly during fast target movements, an improved YoloV7 re-detection algorithm is introduced. When tracking fails, this improved YoloV7 algorithm performs target detection on the current frame and quickly updates the tracking box to achieve target repositioning. Finally, through experimental comparisons of various object tracking algorithms, this approach demonstrates significant improvements in tracking accuracy, precision, and frame rate.