Enhancing Infant Safety: Performance Analysis of Deep Learning Method on Development Board for Real-Time Monitoring

Abstract

Supervising an infant while managing household tasks poses a significant challenge for parents, often leading to safety concerns due to limited time and attention for constant monitoring. Conventional sensor-based monitoring systems have limitations in accurately detecting an infant’s condition and position. This research proposes a deep learning algorithm method as an alternative to sensors. The deep learning algorithm will be integrated with a camera to assess the infant’s position and condition using the Convolutional Neural Network (CNN) method. To enhance performance, a development board integrated with a Video Graphic Array (VGA) is utilized to accelerate the processing time of the deep learning method, and the CNN model is simplified using the MobileNetv2 architecture to reduce model weight. Testing with various scenarios, including an infant in potentially hazardous positions (i.e., prone, covered by an obstacle, or standing near a fence) and an infant in safe positions, achieved an accuracy of 94%. This indicates that the model can effectively identify the infant’s position and determine their well-being. Additionally, the Frame Per Second (FPS) results of 20 demonstrate real-time infant supervision capabilities, allowing for timely intervention in case of any potential risks. Based on the experimental findings, this method holds promise for applications in the medical field and for parents who may struggle with continuous infant supervision, enabling them to fulfill other responsibilities without compromising infant safety.