Teknomo–Fernandez Kernelized Discriminant Analysis-Based Connectionist Deep Multilayer Perceptive Neural Learning for Human Activity Recognition
R. Bagavathi Lakshmi, A. Saranya · Apple Academic Press eBooks · 2024
Human activity recognition is one of the upcoming types of research in recent days for identification of the movement or action of a person depending on variety of sensor data. Although much of the work has been done on human activity recognition but accurate recognition is the main issue due to the complexity and uncertainty of real scenarios. A novel technique called Teknomo–Fernandez kernelized discriminant Feature based Connectionist deep multilayer perceptive neural learning (TFKDF-CDMPNL) has improved the accuracy in human activity detection to a great extent. The activity is monitored by fitting the wearable sensors in the human body and then storing the dataset pertaining to human activity. The dataset is multivariate and neural networks are used to train the model. Basically, the Multilayer Perceptron with input, 114 hidden, and output layer is used. Multiple video sequences are taken as input-by-input layer and then transferred to first hidden layer where they are packaged into frames. Teknomo–Fernandez algorithm determines foreground and background frames after separating them. After that the Radial basis kernelized discriminant Feature extraction process is carried out for finding important and robust features so that the process of human activity recognition be done in optimal time. The last layer is used for classification and Czekanowski dice similarity coefficient is used to analyze features and final classification. The similarity value is studied by using binary step activation function with threshold value. The results shown by TFKDF-CDMPML have been remarkably accurate for human activity recognition activity. Accuracy, precision, recall, F-measure, and time complexity with respect to the number of video sequences are some of the performance measures used to check the performance and based on the results the proposed TFKDF-CDMPNL technique has shown improvement in the performance of human action recognition accuracy with lesser time consumption.