Fall Detection by a SVM-Based Cloud System with Motion Sensors

Recently, fall detection has become a popular research topic to take care of the increasing aging population. Many previous works used cameras, accelerometers and gyroscopes as sensor devices to collect motion data of human beings and then to distinguish falls from other normal behaviors of human beings. However, these techniques encountered some challenges such as privacy, accuracy, convenience and data-processing time. In this paper, a motion sensor which can compress motion data into skeleton points effectively meanwhile providing privacy and convenience are chosen as the sensor devices for detecting falls. Furthermore, to achieve high accuracy of fall detection, support vector machine (SVM) is employed in the proposed cloud system. Experimental results show that, under the best setting, the accuracy of our fall-detection SVM model can be greater than 99.90 %. In addition, the detection time of falls only takes less than 10⁻³ s. Therefore, the proposed SVM-based cloud system with motion sensors successfully enables fall detection at real time with high accuracy.