Machine learning and uLBP histograms for posture recognition of dependent people via Big Data Hadoop and Spark platform

Abstract

For dependent population, falls accident are a serious health issue, particularly in a situation of pandemic saturation of health structures. It is, therefore, highly desirable to quarantine patients at home, in order to avoid the spread of contagious diseases. A dedicated surveillance system at home may become an urgent need in order to improve the patients’ living autonomy and significantly reduce assistance costs while preserving their privacy and intimacy. The domestic fall accident is regarded as an abrupt pose transition. Accordingly, normal human postures have to be recognized first. To this end, we proposed a novel big data scalable method for posture recognition using uniform local binary pattern (uLBP) histograms for pattern extraction. Instead of saving the pixels of the entire image, only the patterns were kept for the identification of human postures. By doing so, we tried to preserve people’s intimacy, which is very important in ehealth. To our knowledge, our work is the first to use this approach in a big data platform context for fall event detection while using Random Forest instead of complex deep learning methods. Application results of our conduct are very interesting in comparison to complex architectures such as convolutional deep neural networks (CNN) and feedforward deep neural networks (DFFNN).