Classification of human motions using kinect sensor

In recent years, studies have been carried out to classify human movements in many areas such as health and safety. Image processing and classification algorithms, which have been used more in recent years, have started to be used in this field as well. The predecessor methods of these classification algorithms are machine learning and deep learning. In this study, the classification of human movements was made on the data set obtained using the Kinect sensor. It is the CAD60 dataset, which is available in the literature, which contains real-time human posture information and images in the data set used. In this dataset, there are data containing different movements/stances of different people. Within the scope of this study, human movements were classified using deep learning-based and machine learning-based methods using MATLAB application. Classification studies were carried out by applying machine learning-based methods to the data obtained by backward feature selection (BFS) and feature extraction with Long Short Term Memory (LSTM) on the raw data set. In addition, human movements are classified with LSTM, Convolutional Neural Networks (CNN), learn from scratch and transfer learning method. The success values obtained at the end of these six methods were compared with each other and the maximum success value was obtained in the feature extraction method with LSTM.