Design and Implementation of an IoT and TinyML-Based Child Safety Monitoring System for School-Aged Children

Abstract

Due to the alarming rise of child abuse, kidnapping, and neglect in regions of West, Central, and East Africa, where rates of violence are among the highest globally, children face major threats to their safety and well-being. Kenya is one of the countries where such safety issues are prevalent. In busy cities like Nairobi, children face many threats in urban settings, especially when commuting to or from school. Many cases of violence against children have been occurring over the past few years, and school children also sometimes fall sick and lack efficient care in case of an emergency. Child kidnapping, abuse, neglect, and fires at schools are the most common concerns, and thus, parents are always concerned about the well-being of their children when they are away. Schoolgoing children are particularly vulnerable to these threats during their transit to and from school, particularly in public transport, crowded areas, and unsupervised school hours. All these risks call for a prompt response, which may not always be available to them. Therefore, a real-time child safety monitoring system is paramount to offer better protection measures in the modern world. This research aimed to design and develop an IoT and TinyML-enabled child safety monitoring system capable of continuously monitoring vital signs of temperature and heart rate, and location tracking via GPS, through a wearable device. With further incorporation of sound sensors, the system detects screams and sends immediate SMS alerts to the parents' mobile devices. The TinyML algorithms, by on-device processing, can identify abnormalities and send real-time alerts to parents in case of emergencies. When in danger or distress, the child can also press a panic button, thus allowing for a proactive response. Most current child protection and supervision technologies and methods are resource-heavy, expensive, and not applicable in resource-constrained scenarios due to their reliance on extensive infrastructure. During this study, data was collected and analyzed to provide insight. Other steps included prototyping, model training, and testing. The development and evaluation procedures used in assessing the effectiveness and performance of the system as part of the study technique involved the design of a wearable IoT device, development and training of TinyML model for sound event detection and testing. This solution was tested with real-time data and reflected high accuracy in identifying emergencies. The study found that the system, using convolutional neural vii networks, gives an 80.8% accuracy in scream detection with low power consumption, hence suitable even for low-resource environments to enhance child safety in rural and urban settings. This demonstrates the potential of IoT and TinyML in transforming child safety monitoring systems. The study contributes to the existing child safety research by demonstrating the potential of neural networks in TinyML for low-power and real-time monitoring solutions in resourceconstrained environments. This work demonstrates a feasible approach to significantly enhance child safety, offering more effective protection measures