International Journal of Computer Techniques Volume 12 Issue 4 | Advancing Public Activity Recognition in Video Streams Using Hybrid Deep Learning Techniques: A Review
Advancing Public Activity Recognition in Video Streams Using Hybrid Deep Learning Techniques: A Review
International Journal of Computer Techniques – Volume 12 Issue 4, July – August 2025
ISSN: 2394-2231 | https://ijctjournal.org
Abstract
This review explores hybrid deep learning techniques for public activity recognition in video streams, a critical component of surveillance and smart city infrastructure. It analyzes CNN-RNN combinations, graph convolutional networks, and transformer-based models, comparing their performance across datasets and metrics. The paper identifies key challenges such as data scarcity, computational overhead, and privacy concerns, and proposes future directions including self-supervised learning and ethical AI frameworks.
Keywords
Public Activity Recognition, Video Streams, Hybrid Deep Learning, CNN, RNN, Surveillance
Conclusion
Hybrid deep learning models have revolutionized public activity recognition by integrating spatial and temporal dynamics. While CNN-RNN, GCN, and transformer-based architectures offer superior accuracy, challenges like annotation cost, scalability, and ethical deployment remain. Future research should focus on efficient, privacy-preserving, and generalizable models to enable real-time, responsible surveillance systems in public spaces.
References
- Dosovitskiy et al., “Transformers for Image Recognition,” ICLR, 2021.
- Howard et al., “MobileNets,” arXiv:1704.04861, 2017.
- Feichtenhofer et al., “SlowFast Networks,” ICCV, 2019.
- Dwork & Roth, “Differential Privacy,” Foundations and Trends, 2014.
- Chen et al., “Temporal Fusion for Activity Recognition,” IEEE TCSVT, 2022.
- Carreira & Zisserman, “Kinetics Dataset,” CVPR, 2017.
- Simonyan & Zisserman, “Two-Stream Networks,” NIPS, 2014.
- Sabokrou et al., “Deep-Anomaly Detection,” CVIU, 2018.
- Misra et al., “Self-Supervised Video Re-ID,” CVPR, 2020.
- Yan et al., “ST-GCN for Skeleton Action Recognition,” AAAI, 2018.
Post Comment