Fall Detection from Words and Motion: A Multimodal Approach to Elderly Safety

Abstract

Elderly falls are a serious public health concern because they frequently cause serious injuries. Because they allow for quick assistance, fall detection devices have become popular ways to lessen the impact of such events. During a fall, elderly people frequently scream or ask for assistance in their native tongue. A multimodal strategy can significantly increase detection accuracy and provide prompt response in emergency situations by combining voice distress signals with fall detection sensors. The proposed system is designed to recognize distress voice patterns in Malayalam, the official language of Kerala. It integrates voice recognition with fall detection to accurately identify emergencies, send SMS alerts to predefined contacts, and activate a buzzer to prompt immediate assistance. Accelerometers are used as the primary sensors for fall detection, and periodic sampling of the sensor data makes the system more energy efficient. Additionally, the Malayalam voice recognition module verifies the emergency by recognizing distress calls, thereby enhancing system reliability and minimizing false positives.