<![CDATA[The increase in student density in educational institutions has generated greater risks to child safety, especially in high-traffic scenarios where manual supervision is limited to prevent accidents. In this context, the need arises to implement non-invasive technologies and systems based on artificial vision, to improve monitoring and support traditional surveillance methods. Previous research describes the use of computer vision techniques and models such as YOLO for safety, monitoring, and their integration into low-cost embedded systems, with limitations in real-time response or accident prevention. This research shows the development of a monitoring system based on artificial vision with a YOLOv8 model trained and deployed on a Raspberry Pi with integration with an ESP32 to detect risk events aimed at preventing accidents in minors. The results indicate that risk classes are detected with confidence levels above 70%, with good performance in metrics such as precision (~0.9), recall (~0.8) and mAP50 (~0.9), without overfitting and with hypothesis validation. In conclusion, the system is a suitable solution for accident detection with metric validation, but it needs to be optimized to work in real-world heterogeneous scenarios.]]>
Copyrights © 2026
