Best Feature for CNN Classification of Human Activity Using IOT Network
Hamdi Amroun, M’hamed Temkit, Mehdi Ammi · 2017
In this paper, we design an experience that evaluates what is the best descriptor to recognize human activity using Convolutional Neural Network (CNN) in a non-controlled environment using a network of smart objects. We chose to classify four types of activities: standing, sitting, laying and walking. We selected a set of the most popular and suitable descriptors and do a comparative study of the classification results using different classifiers. Results show that the discrete cosine transform (DCT), with the convolutional neural network (CNN) as a classifier, achieves more than 98% average accuracy by choosing a certain provision of the network of smart objects on the body. Therefore, the selection of descriptors saves computation time and memory space, reaching high classification accuracy.