A Systematic Review of Multi-Object Tracking Using Deep Learning in Video Sequences
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Multi-object tracking in video sequences is discussed as one of the most fundamental tasks in the computer vision. It is used in a very diverse set of applications, such as autonomous driving, sports analytics, surveillance, and human-computer interactions, while acknowledging persistent challenges like occlusions, appearance variations, and real-time demands. Through the advancements made in the field of deep learning, recent years have seen a radical change in this field, allowing end-to-end learning, feature extractions that are more powerful, and more accurate tracking, among other benefits. The paper illuminates the latest state of art in the domain of deep learning algorithms in multi-object tracking systems, focusing on its approaches, performance, and constraints. It examines detection-based and joint detection-tracking frameworks, evaluating their performance on benchmarks such as MOT Challenge and KITTI using metrics like MOTA and MOTP. The review highlights ongoing limitations in real-time processing, occlusion handling, and generalization across environments. To address these, it proposes future research directions, including multimodal data integration, self-supervised learning, and interpretable AI, aiming to overcome current system constraints and improve robustness.
Fields of Study & Subjects
Video Surveillance and Tracking Methods
Autonomous Vehicle Technology and Safety
Advanced Neural Network Applications
Publication details
- Pages
- 1–15
- Publication date
- 1 July 2026
- Language
- EN