Active Human Pose Estimation for Assisted Living

Abstract

Active and Assisted Living has found itself in one of the application areas of technological advancement the world is witnessing. The objective is to provide the elderly people with facilitated living environment so as to assist them in carrying out daily activities without them being prone to injury or any other undesirable event. These residential facilities prove to be even more beneficial when equipped with technology to prevent any fatalities or aid immediately in case fatality occurs. One such harmful event is falling. Falling especially in the case of elderly can have serious impacts on their health. Hence, several attempts have been made to provide aid immediately whenever such event occurs. This includes usage of different techniques like Wearable sensors, Computer vision or Ambient sensors. This paper aims at exploring Computer vision technique to determine fall. For this, key points of human body are located which are then used to identify if the fall has occurred or not. The proposed algorithm uses publicly available dataset to train on detecting fall. Several classifiers like SVM, AdaBoost, Logistic Regression has been used for classification with SVM reporting 82.07% accuracy, AdaBoost with 99.64% accuracy and Logistic Regression with 98.92% accuracy.