Year: 2024
Type: conference-paper
Abstract: Nowadays, researchers' studies on target tracking have become one of the important research directions in computer vision, and various target tracking models with excellent characteristics have been proposed continuously with the depth of research. The results of these models in real-world target tracking applications, however, do not always maintain their robustness, due to the fact that the tracking process can be affected by factors such as frame occlusion, illumination variations, and fast motion. In this paper, the performance of the deep learning-based SiamRPN (Siamese-RPN) model for target tracking is deeply investigated, and the CBAM (Convolutional Block Attention Module) mechanism is added to the twin network structure of this model to compensate for the shortcomings of the original model. The improved model is proved to be indeed much more accurate and precise for target recognition than the original model by comparing the experimental data.
Author: Yu Wang
Institution: Chengyi College, Jimei University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4409965874