Multi-Person Detection and Re-Identification in Surveillance Systems Using Deep Learning

This paper has presented a complete, deploy-able pipeline for multi-person detection and re-identification in surveillance video.  By fusing a metric-learning ResNet-50 embedding with color, texture and body-proportion cues into a single 698-dimensional descriptor, and complementing that with a face verification side-channel built on face-cue and InsightFace ArcFace embeddings, and by assign-ing identities through an ordered multi-pass engine with a co-occurrence safety net, the system is able to hold identity labels steady through occlusion, similar-looking clothing and temporary exits from the scene. Evaluation on recorded footage and on the Market-1501 benchmark reports a person-class [email protected] of 0.93, a Rank-1 accuracy of 0.87, an identity recovery rate of 0.89 and an identity switch rate below 5% at 2–3 frames per second on CPU, with projected throughput of 18–22 frames per second on GPU hardware. Future work will explore a vision-transformer backbone to capture longer-range spatial depen-dencies, domain-specific fine-tuning directly on the recorded surveillance dataset, direct RTSP/IP-camera stream ingestion (the current live path only accepts frames from a browser webcam rather than network cameras), ONNX-based quantisation for edge deploy-ment on devices such as the Jetson Nano, and fusing in gait or skeleton-derived features to better tell apart individuals wearing similar clothing.

[1]     M. Ye, J. Shen, G. Lin, T. Xiang, L. Shao, and S. C. H. Hoi, “Deep learning for person re-identification: A survey and out-look,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 44, no. 6, pp. 2872–2893, 2021. [2]     K. Zhou, Y. Yang, A. Cavallaro, and T. Xiang, “Omni-scale feature learning for person re-identification,” in Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV), 2019, pp. 3702–3712. [3]     D. Mendes, R. Cerqueira, S. Avila, and D. R. da Silva, “Multi-camera person re-identification based on trajectory data,” Ap-plied Sciences, vol. 13, no. 8, 2023. [4]     M. V. Kumar, A. Arulmurugan, and B. R. S. Ram, “Deep learning based multi-camera person tracking and re-identification system,” Int. J. Multidisciplinary Research, vol. 6, no. 1, 2024. [5]     O. Lachihab, S. Lachihab, and B. Bouchama, “Fine-tuning YOLOv8s for unified human and face detection in crowded environments,” Eng. Technol. Appl. Sci. Res., vol. 16, no. 1,pp. 32348–32356, 2026. [6]     L. Zheng, Y. Yang, and A. G. Hauptmann, “Person re-identification: Past, present and future,” arXiv preprint arXiv:1610.02984, 2016. [7]     D. Tao, Y. Guo, M. Song, Y. Li, F. Yu, and Y. Shao, “Deep multi-view feature learning for person re-identification,” IEEE Trans. Circuits Syst. Video Technol., vol. 28, no. 10, pp. 2637–2646, 2018. [8]     E. Ahmed, M. Jones, and T. K. Marks, “An improved deep learning architecture for person re-identification,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2015,pp. 3908–3916. [9]     S. Saber, S. K. Srivastava, and S. K. Singh, “A multi-attention approach for person re-identification using deep learning,” Sensors, vol. 23, no. 7, p. 3678, 2023. [10]     J. J. M. Borjon, G. J. L. Herrera, and C. A. R. Garay, “Us-ing deep learning embeddings for the person re-identification problem,” in Proc. IEEE Int. Autumn Meeting Power, Elec-tron., Comput. (ROPEC), 2025.