Context-Aware Complex Human Activity Recognition Using Hybrid Deep Learning Models

Smart devices, such as smartphones, smartwatches, etc., are examples of promising platforms for automatic recognition human activities. However, it is difficult to accurately monitor complex activities on these due interclass pattern similarities, which occur when different exhibit similar signal patterns or characteristics. Current smartphone-based systems depend traditional sensors, accelerometers and gyroscopes, built-in in devices. Therefore, apart from using information the lack contextual support activity recognition. In this article, we explore environmental contexts, illumination (light conditions) noise level, sensory data obtained sensors a hybrid Convolutional Neural Network Long Short-Term Memory (CNN–LSTM) learning models. The models performed sensor fusion by augmenting low-level signals with rich improve models’ accuracy generalization. Two sets experiments were validate proposed solution. first set used triaxial inertial sensing train baseline models, while second combined sensors. results demonstrate that information, level light conditions deep achieved better than without information.