A multimodal sensing system for elder fall detection

Abstract

Among the many health care issues for elder people, fall detection has shown up as primary focus. We propose a sensor fusion concept for elder fall monitoring and analysis. The idea is to integrate depth sensor and motion sensor. We illustrate the sensor fusion concept and implement the wireless fall detection sensor sub-system. We experiment with angular acceleration data in three limbs parts of human test target. Different life patterns of elderly are selected to test the motion monitoring system: walking and falling. Two methods are employed as diagnosis tool. The first is the energy expenditure method, while the second is the neural network method. The first tests the sensitivity of motion sensor, while the second filters out the noisy motion data for correct classification purpose. We discovered that the trained weights and bios from the neural network based on three limbs parts measurement data has allowed us to differentiate fall from regular walk motion with high accuracy.