Machine Vision-Based Fall Detection System using MediaPipe Pose with IoT Monitoring and Alarm

The incidence of falls is more commonly experienced by the ageing population globally, due to their increasing frailty. Fall detection systems are being developed combined with machine learning approaches that include those wearable devices, ambience-based systems, and vision-based systems. Wearable systems are omnipresent but intrusive while ambience-based systems are temperature-dependent that require a dedicated GPU. Challenges on vision-based fall detection systems include large computational cost, use of specialized camera, and system integration into smartphones. This research aims to propose a vision-based fall detection system ported on a smartphone application, that utilizes deep learning trained and tested from multiple RGB camera setups. In an event of a fall, the system is designed to provide an on-premises auditory alarm, an IoT notification, and a real-time video feed via the smartphone application. Using the pretrained MobileNetv2 CNN-based MediaPipe Pose and Random Forest Classifier for fall detection, experimental results show high-performance evaluation metrics.