RE-IDENTIFICATION PROBLEM FOR OBJECTS WITH INDEPENDENT TRAJECTORIES
Received: 2025-05-20
Published: 2026-06-06
Abstract
Multi-Object Tracking (MOT) systems are one of the important directions of modern computer vision and artificial intelligence, designed to detect and track several moving objects in real time. MOT systems enable not only object detection but also identification across consecutive video frames over time. Nevertheless, their accuracy and reliability are limited by a number of technical problems. In particular, occlusion, when one object is temporarily hidden by another object or the background, can reduce tracking accuracy and cause identity switches or loss of a target. Objects may also temporarily leave the camera field of view, while illumination, motion speed, and complex backgrounds can negatively affect tracking. To address these problems, modern MOT systems use deep learning, convolutional neural networks (CNN), recurrent neural networks (RNN), transformer architectures, and object re-identification (Re-ID) technologies. Models for predicting object trajectories and restoring temporarily lost targets are especially important.
Keywords
List of references
-
Qiankun Liua,1, Dongdong Chenb,1, Qi Chua,∗, Lu Yuanb, Bin Liua, Lei Zhangc, Nenghai Yua. Online Multi-Object Tracking with Unsupervised Re-Identification Learning and Occlusion Estimation. January 4, 2022.
-
Babaee, M., Li, Z., Rigoll, G., 2019. A dual cnn–rnn for multiple people tracking. Neurocomputing 368, 69–83.
-
Bergmann, P., Meinhardt, T., Leal-Taixe, L., 2019. Tracking without bells and whistles, in: IEEE International Conference on Computer Vision, pp. 941–951
About the Authors
License
This work is licensed under a Creative Commons Attribution 4.0 International License.