What is it about?

With the evolution of technology and robotic systems, unmanned aerial vehicles (UAVs) are increasingly exploited in real-world applications. For this achievement computer vision and artificial intelligence methodologies are required to visually extract information about the operational environment. A challenge of this topic is the development of appropriate methods that extract robust information and operate fast under limited computational resources. These concerns are the playground of this work, which presents a benchmarking of visual human tracking algorithms in images acquired by a camera mounted on a UAV. More specifically 37 tracking algorithms have been benchmarked under performance and efficiency.

Featured Image

Why is it important?

The contribution of this work is a performance benchmarking of human tracking algorithms under various metrics that describe the performance of the visual tracking problem and the computational resources' footprints. More specifically the novelty of the proposed paper is a systematical benchmarking of lightweight tracking algorithms with a wide range of factors around problem solutions and device resource consumption. For the first time, human tracking algorithms are massively benchmarked in new visual object tracking datasets. From the results, deep learning trackers are performing much better than correlation filter trackers, achieving a better balance between performance and efficiency.

Perspectives

It was a great pleasure to implement this benchmarking study and collaborate the authors to produce this article. Through our joint efforts, we were able to provide a thorough evaluation of tracking algorithms under challenging circumstances, which we believe will be a valuable resource for researchers and practitioners in the fields of UAVs and computer vision applications. Our hope is that this article will raise awareness and generate interest in these areas, encouraging further exploration and innovation. By providing comprehensive knowledge about tracking algorithms and their performance under various conditions, we aim to facilitate the development of more effective and efficient solutions for these important applications.

Theofanis Kalampokas

Read the Original

This page is a summary of: Performance Benchmarking of Visual Human Tracking Algorithms for UAVs, November 2022, ACM (Association for Computing Machinery),
DOI: 10.1145/3575879.3575880.
You can read the full text:

Read

Resources

Contributors

The following have contributed to this page