Activity Recognition System via Unification of CNN and SVM in Complex Domain

Human automatic activity recognition is an essential part of a little interactive application which involves human-being. The main disadvantage is the person’s diverse activities. These techniques provide high accuracy and pattern recognition. Deep learning methods succeed in human activity recognition (HAR); hence our goal is to combine Convolutional Neural Networks (CNN), which work with image data and extract features from images with Support Vector Machines (SVM), a shallow architecture that maps low-dimensional space to high-dimensional space. This paper uses an intelligent system algorithm to detect human activity and recognize its pattern. Cooking, cleaning, dancing, steering, and discussion activities are to be tested. Each class has 7480 images. Applying all preprocessing techniques, experimental results shows that the cleaning class gives the highest training and testing accuracy i.e., 98.78% and 97.51%. This method achieves the highest recognition accuracy with the lowest computational cost.