A Literature Survey of Human Activity Recognition Using Deep Learning and Nonparametric Model with Some Exchanges in Karl Popper’s Viewpoint and Kuhn’s Paradigm in Philosophy of Science

Abstract

Human skeletal detection and human gesture recognition are interesting subjects that have been investigated during the past three decades. Single-RGB, RGB-D camera, and Initial Measurement Unit (IMU) are some of the sensors for recording human motion data. Numerous methods for gesture recognition and classification have been reviewed in this survey. The classification is divided into nonparametric models and deep learning models, which afterwards will be compared in terms of accuracy and running time, respectively. The feature extractions are separated based on features processed from the sensor data, including skeleton-based features, depth image-based features, and hybrid features. A comparison of accuracy values will be offered based on the model and its attributes. In addition, we present an interchange of perspectives on deep learning and nonparametric models based on Karl Popper’s perspective and Kuhn’s paradigm in the study of the philosophy of science. By substituting the falsification principle for induction, Popper attempts to refute the traditional empiricist perspective of the scientific method. From the philosophy of science’s perspective, the study on human action recognition is in the normal science phase according to Kuhn’s paradigm and is corroborated in accordance with Popper’s theory.