Peer-reviewed article
Vision-based system model for detecting violence against children
Samir Hammami, Muhammad Alhammami ·
Summary
Written by AI, not yet AICS's view · claude-haiku-4-5-20251001This paper describes a machine learning model designed to detect violence against children using skeletal data from depth sensors. The model achieved 99.03% accuracy in classifying violent versus non-violent actions while protecting privacy by using only skeleton joint information. The authors suggest the technology could be implemented in schools by psychologists and counselors.
Key points
- Machine learning model achieves 99.03% accuracy in detecting violent actions against children
- Uses only skeletal data from depth sensors, preserving privacy compared to video-based approaches
- Binary classification system: violent action or non-violent action
- Authors propose implementation potential in school settings with psychologists and counselors
Limitations
This is a technical methods paper that describes a model's performance on test data but provides no evidence of real-world effectiveness, validation with actual child protection practitioners, or evaluation of the intervention's impact on children's safety or well-being. It does not evaluate whether the system actually improves detection or outcomes for children experiencing violence.
Abstract
We present in this paper a machine learning model for detecting violence against children. This model, which uses skeletal data acquired by depth sensors achieved a high accuracy violence detection rate of 99.03 %. In sum, this research method presents: •First ML-based method for detecting most common child abuses, which keeps the privacy of people by using only skeleton joints data.•The model has only two classes (violent action, non-violent action).•The model can be a base for other researches and implementations in schools by school psychologists and counselors.