Datasets

Modern research in the field of Human Activity Recognition primarily revolves around Machine Learning and Deep Learning, due to their substantial advantages in terms of adaptability, precision, and enhanced processing speeds compared to conventional methods. Innovating on various aspects of the activity recognition pipeline, several distinctive deep learning algorithms have arisen during the past few decades. The ‘state of the art’ machine learning models now available are frequently quite data-intensive and require a sizable training dataset. Several datasets have been produced in order to meet this need. Depending on the requirements of the algorithm being trained, different datasets may be chosen for use under different conditions. In this chapter a detailed review of researches carried out in evolution of the dataset starting from early age to modern dataset has been carried out. The chapter covers a short description on characteristics of datasets in terms of their focus, modality, data source, Annotation Method, Annotation Type and evolution followed by evolution of modern and early age datasets. Finally, some frequently used datasets along with their short description followed by conclusion has been presented.