Human fall detection using non-contact radar sensor

Abstract

Falling is the main cause of disability and fatality of elderly. In this work, 24 GHz continuous wave Doppler radar is used to develop a low price fall detection system. Radar sensor is selected due to its capability of tracking human motions, passing through covers and walls, its low cost, low power, and small size. Designed system is further improved to detect and monitor the human’s vital signs to analyze the status of the fallen person and reduce the consequences of the fall by providing a general idea about the persons’ situation to the concerned authorities. First, considering all possible daily activities and fall cases, a dataset with 121 fall and 117 non-fall signatures are collected. Then, features from both time and frequency domains are extracted and examined to select the ones that contribute most to distinguish between fall and non-fall samples. Finally, different machine learning techniques including support vector machine, naive Bayes, k nearest neighbor, linear discriminant analysis and decision tree are evaluated to build the most accurate classification model. Proposed system performed activity classification and fall detection with 88% average accuracy, heart rate monitoring with 95% average accuracy, and respiration rate monitoring with 85% average accuracy.