Elderly people activity monitoring System based on ambient assisted Living

Abstract

The medical field advancements have considerably increased the newlinelifespan of elderly people aged over 65. The increase in the lifespan of newlinehumans leads to a high dependency rate which necessitates the need for newlineambient assisted living. Recognizing human activities automatically is very newlineimportant for most ambient intelligence applications. The human activity newlinerecognition system classifies the actions performed by a person using the set newlineof observations. newlineFor the real-time implementation of activity recognition systems, newlinethe processing time of acquired data must be minimal. The feature dimension newlinemust be reduced to decrease the processing time. Secondly, classification newlinetechniques are very important and an efficient classification technique is newlinerequired to classify the activities accurately. Pre-trained networks like newlineAlexNet, can only be used for image domain applications. To use pre-trained newlinemodels for IMU sensor data, the IMU sensor data must be converted to the newlineimage domain. So, an image conversion algorithm can be used for achieving newlinethe task. Further, the object detection methods like YOLO can be used to newlineidentify the activities with 45 frames per second, which is very fast. newlineThe main focus of this research work is on four objectives. The first newlineobjective is to develop a feature dimensionality reduction technique to reduce newlinethe number of features. The second objective is to develop a classification newlinetechnique to efficiently classify the performed activities. The third objective is newlineto develop an image conversion method to convert IMU data to the image newlinedomain to make it compatible with pre-trained networks like AlexNet. The newlinefourth objective is to use the object detection techniques like YOLO for newlineactivity recognition newline