Multi-Object Tracking with Nonlinear Motion Modeling and Discriminative Appearance Features
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 (MOT) is a fundamental computer vision task that establishes target associations across consecutive video frames, playing a critical role in applications like autonomous driving and artificial intelligence. However, MOT remains highly challenging, especially when targets are small, move rapidly, or appear in crowded environments. A representative MOT algorithm, DeepSORT, faces limitations in both accuracy and efficiency due to its reliance on outdated detection and data association modules. This study proposes an enhanced DeepSORT version by improving its two core components: object detection and data association. Specifically, we replace the original Faster R-CNN detector with YOLOE, which achieves significantly higher detection accuracy and faster inference speed. For data association, we introduce a lowconfidence association module and an improved Kalman filter to further boost tracking performance. Furthermore, we incorporate a Fisher Linear Discriminant (FLD) strategy to enhance the discriminability of features for visually similar targets. This strategy effectively increases the accuracy of appearance matching and reduces frequent identity switches among pedestrians. Overall, our improvements substantially enhance the accuracy and robustness of DeepSORT in complex tracking scenarios.
Fields of Study & Subjects
Video Surveillance and Tracking Methods
Autonomous Vehicle Technology and Safety
Advanced Neural Network Applications
Publication details
- Journal / source
- International Conference on Electrical Automation and Artificial Intelligence (ICEAAI)
- Pages
- 804–807
- Publication date
- 16 January 2026
- Language
- EN