Preventing Falls & Accidents of Babies in a Crib Using Machine Learning-Based Intelligent Alert System

Abstract

Babies play a very important role in a healthcare setting. Babies aged 6-24 months are at a crucial stage of growth. They need to be supervised because of their growing age, developmental stage, and motor skills. These are the reasons why babies are in dangerous postures and unsafe scenarios, leading to falls, which sometimes lead to fatal injuries, severe head injuries, bone fractures, etc. Approximately 52% of hospitalizations for falls involve babies aged 0-4 years. Around 31% of fall-related hospitalizations happen at home. Parents who work can't always keep an eye on their newborns, so they use nannies, daycare services, or smart technology to monitor their babies 24/7 and keep them safe. But concerns about caregivers' trustworthiness and the safety of childcare arrangements cause stress and underscore the importance of affordable, reliable e-health solutions that let you always keep an eye on your infant. In modern times, Artificial Intelligence (AI) and Machine Learning (ML) play a very important role in daily life in healthcare applications, serving as a vital companion for babies at home in the crib, providing continuous vigilance of babies' activities, early fall risk predictions, and resolving the issues of negligence.

The focus of this thesis is to identify fall risk scenarios among babies within the crib setting by analyzing behaviour patterns using AI and Machine Learning techniques. The proposed approach, termed the ADL Conditioned Cascaded AI model for potential baby fall risk prediction, employs Pose features (MediaPipe), a Gated Recurrent Unit (GRU), a Soft Spatial Attention mechanism, and a Dual-Stage Trigger mechanism to predict high-risk ADLs that lead to future baby fall risk. In this research, all the models use Activities of Daily Living (ADLs) within the crib, such as sitting, playing, lying down, leaning over the fence, standing near the fence, and climbing over the fence as input parameters. This thesis presents three significant contributions: • A detailed background analysis of baby fall detection, baby fall prediction, and baby safety monitoring was done in terms of behaviour, identifying scenarios within or outside the crib. • The system architecture mainly consists of the AI-3 model, leveraging the techniques and overcoming the flaws of the AI-1 and AI-2 models to predict the high-risk scenarios leading to falls and generate a more reliable outcome. • Data is collected from online platforms for all the models, and a few videos are generated specifically by an AI tool to test the AI-2 and AI-3 models on an unseen dataset. By testing the AI-3 model in the time domain, it effectively predicts high-risk ADLs that lead to future baby fall risk.

This thesis summarizes the research findings and model advancements that underscore the considerable potential of sophisticated deep learning techniques to enhance future predictions of babies’ fall risk, establish an evolving warning system, and ultimately, forecast and mitigate fall incidents. This cutting-edge technology offers a revolutionary solution that enables prompt interventions, making babies in the crib safer and healthier while also relieving parents of the stress of constant vigilance.