Scopus Indexed Publications

Paper Details


Title
From Detection to Dynamics: An Interpretable Framework for Children's Activity and Social Cluster Analysis in Cctv Footage

Author
, Sabit Ahamed Preanto,

Email

Abstract

Automated behavior analysis in the CCTV environment of the school is difficult because of the high-density crowd, children wearing uniforms, occlusions, and identity instability in multi-object tracking. The appearance-based re-identification approach suffers from the “Person-in-Uniform” problem, which limits the approach's robustness and interpretability. In this paper, we introduce a spatio-temporal kinematic framework for children's activity and social cluster analysis. A fine-tuned YOLOv8m detector is used in conjunction with the BoT-SORT algorithm for identity-aware tracking. The model achieved an mAP@50 of 0.921 on a self-collected school playground CCTV dataset. Instead of relying on computationally expensive deep learning architectures for action recognition, the system categorizes micro-activities such as idle, walk, and run using statistically derived velocity thresholds. Social interactions are modeled using a graph adjacency matrix for the detection of clustering anomalies. Validation of the system shows that the system can track objects effectively and has 85% classification accuracy for micro-activities without relying on labeled datasets for action recognition. The findings illustrate that physics-inspired motion modeling combined with modern detection frameworks provides an efficient and scalable solution for real-time school safety monitoring.


Keywords

Journal or Conference Name
2026 International Conference on Signal Image Processing and Communication, ICSIPC 2026

Publication Year
2026

Indexing
scopus