An Image-based Human Physical Activities Recognition in an Indoor Environment

In this paper, we propose real-time image-based recognition of human activities from series of images considering different human actions performed in an indoor environment.The proposed image-based human activity recognition(IHAR)system can be utilized for assisting the life of disabled persons, surveillance and human tracking, human computer interaction,and efficient resource utilization. The proposed IHAR system consists of closed-circuit television (CCTV) camera based image acquisitioning, various filtering based image enhancement, principle component analysis(PCA) based features extraction, and various machine learning algorithms for recognition accuracy performance comparison. We collected dataset of 10 different activities such as walking, sitting down and standing up consists of 35,530 images. The dataset is divided into(90%,10%),(80%,20%), and(70%,30%)training and testing respectively and evaluated three classifier K-nearest neighbors (KNN), Random Forest (RF), and Decision Tree(DT). The experimental results show the accuracy of 95%, 97%, and 90% by KNN, RF, and DT respectively.