| Abstract |
In recent years, computer vision applications based on deep learning have developed rapidly, and systems such as smart buildings, smart cities and surveillance and security have become key research areas for research scholars. Although scholars at home and abroad have made remarkable achievements in the field of multi-object tracking, in complex scenes with dense targets and indistinguishable target backgrounds, multi-object tracking algorithms still suffer from problems such as target miss detection and false detection, resulting in a large number of disconnected, erroneous and other low-quality target tracking trajectories. To address the above problems, this paper proposes a Multi-Object Tracking Algorithm Based on Multi-layer Feature Adaptive Fusion (MFAF-MOT). This algorithm designs a high-level semantic adaptive weighting module and effectively achieve multi-layer feature fusion. By adaptively weighting deep high-level semantic features and shallow fine-grained features, this algorithm improves the network's ability to express target features and reduce errors such as target miss detection and false detection. The algorithm proposed in this paper achieves competitive results in three different scenarios, reducing problems such as disconnected target trajectories and wrong target trajectories, and improving the precision and stability of the multi-object tracking algorithm. © 2022 IEEE. |
