Person re-identification in images based on a conditional hypermodel

Abstract

This paper provides a systematic review of modern adaptive hypermodel methods for the task of person re-identification. The analysis is based on comparing the effectiveness of models tested in both controlled laboratory conditions and realworld scenarios. Considerable attention is paid to key performance indicators—recognition accuracy (mAP, CMC) and data processing speed—which enables a comprehensive assessment of the methods under study. The focus of the research is the study of the impact of dynamic parameter changes on system performance, as well as the analysis of incremental learning strategies aimed at minimizing the risk of catastrophic forgetting during adaptation to new conditions without the need for full retraining. This approach ensures a rapid response to variations in shooting conditions—changes in lighting, angles, and other characteristics—which is critically important for video surveillance. Based on the obtained results, promising directions for further research are identified, including optimization of adaptive learning algorithms, design of new architectural schemes, and scaling of systems while maintaining performance. The implementation of these ideas will contribute to the creation of more reliable and efficient solutions for re-identification in modern information platforms. A hybrid architecture is proposed that combines a conditional hypermodel with a base deep neural network. The main advantages of this approach lie in high adaptability and training stability, achieved through dynamic adjustment of parameters via the hypermodel. The combination of loss functions—cross-entropy and triplet—contributes to forming compact and well-separated features for different identities, enhancing the model’s ability to correctly identify subjects under high variability of input data. Experimental results confirm the feasibility of integrating adaptive mechanisms into re-identification systems, providing enhanced robustness to environmental changes and the high performance necessary for successful implementation in information technologies.