A Novel Framework to Recognize Complex Human Activity

Human activity recognition has importance in many research areas including healthcare, elderly care, assisted living, and context-aware applications. There have been many researches involving simple human activities like walking, sitting, and running. There are only a few studies which considered recognizing complex human activities such as reading book, watching TV, doing dishes, and cooking. This study presents a novel framework to recognize complex activities. The framework uses the time, location, and simple activity to recognize complex activities. These inputs of the framework have been used to model the observation and the complex activities as the states of a Hidden Markov Model. Then the Viterbi algorithm has been used to find the hidden states (complex activities) from the observations (time, location, and simple activity). Data was collected from three subjects for a total of 56 days consisting 51 unique complex activities. Out-of-sample experimental results show that the proposed approach is able to recognize these 51 unique complex activities. The approach achieved an accuracy of above 97% for all the subjects. The proposed approach leverages time, location, and simple activity information to recognize complex activities. It also achieved upto 94% accuracy with just time and simple activity information.