Performance Analysis of K-ELM Classifiers with the State-of-Art Classifiers for Human Action Recognition

Abstract

Recent advances in computer vision have drawn much attention toward human activity recognition (HAR) for numerous applications similar to video games, robotics, content recovery, video surveillance, etc. The enlightening and pursuing of human actions recognized by the wearable sensor devices (WSD) generally used today face difficulty in precision and reckless automatic recognition due to regular change of body movements by the human. Primarily, the HAR system will preprocess the WSD signal, and then, six sets of features were extracted from wearable sensor accelerometer data that are viable from the computational viewpoint. In the end, after the crucial dimensionality reduction process, the selected features were utilized by the classifier to ensure high human action classification results. In this paper, to analyze the performance of the K-ELM, classifiers-based deep model for selected features is predominantly focused with the state-of-the-art classifiers such as artificial neural network (ANN), k-nearest neighbor (KNN), support vector machines (SVM) and convolutional neural network (CNN). The experimental results obtained by analyzing performance using the metrics such as Precision, Recall, F-measure, specificity and accuracy shows that K-ELM outperforms with less time for most of the above-mentioned state-of-the-art classifiers.