Exer-NN: Cnn Based Human Exercise Pose Classification

Abstract

To enjoy the glow of good health, you must exercise [Gene Tunney], because it helps us to feel happier, increase energy levels, reduce chronic disease and helps us to keep our brain and body refresh. Technology can make an important role to help aspects of our life in different portions. Today's computer vision technology uses deep learning algorithms that use the so-called convolutional neural networks (CNN), make sense of pictures. We can use the convolutional neural network (CNN) in deep learning to get state-of-art accuracy in different in various classification problems like as Image data, CIFAR-100, CIFAR-10, MINIST data sets. In this work, we propose a novel system to classify different types of human exercise pose detections automatic self-ruling decision making and predictive models using Convolutional neural networks (CNN). In earlier a lot of research has been conducted to pose detections in image classification problems, but our related tropic human exercise pose detection problem has few works on different data sets and different models with low accuracy. For strong architecture, we retrained the final layer of the CNN architecture, VGG16, MobileNet, Inception V3 for classification approach. We will create a new CNN model name ‘Exer-NN’ to successfully classify human exercise pose. Predicting among five different classes (bench press, bicep curl, squat, deadlift, treadmills). We proposed an average accuracy is 88% approximately that can be used for different purposes like tool kit assistance, helping management system automatically.