Survey on approaches and practical areas of human activity recognition application

Human activity recognition is one of the relevant fields of research in machine learning since recognition results are necessary for solving many practical problems. The article provides a survey on approaches and areas of practical application of methods for human activity recognition. The sensors used for human activity recognition are considered, and the criteria for their selection are presented. Possible solutions to the problem of choosing the location and orientation of wearable sensors are presented. The main stages of human activity recognition are discussed in the article. Extracted features and methods of their selection to increase the accurate classification of human activity recognition and reduce computational complexity by cutting down the number of features are presented. The advantages and disadvantages of popular classification methods are formulated. The metrics used to evaluate the quality of learning classification models are considered. The most commonly used quality metric is the error curve. Practical tasks in which the results of human activity recognition are needed are also presented. The main areas of human activity recognition application are medicine, manufacturing, fitness, and safety of people. In conclusion, possible future research directions are presented.