Deep learning-based tools and communications for sensing

The advancement of deep-learning technologies and the pervasive deployment of communication networks open new possibilities for sensing applications. Deep-learning tools are changing the paradigm from passive data collection to enhanced possibilities of interpretation and meaningful pattern extraction. On the other hand, communications are becoming widespread and play a crucial role in our society, thus the integration of sensing features appears as a natural consequence of the maximum exploitation of the communication infrastructures and for the enhancement of communications themselves. In general, sensing is a fundamental and indispensable ability in our daily existence because provides information and allows our interaction with the world, thus the possibilities offered by these relatively new tools have the power to innovate society. In recent years, different domains have been transformed by sensorization as automation in healthcare, industry, agriculture, auto- motive, and others. This thesis contributes to the field of sensing from different research perspectives. First, remote sensing is explored for movement monitoring in the clinical gait analysis context. We developed two multidisciplinary approaches involving computer vision and radar sensing coupled with deep learning processing neural networks for data processing. Then, we focused on multimodal acquisition systems. The information derived from different signals collected simultaneously is considered and enabled. Some practical problems related to the creation of adequate set-ups are discussed and we proposed both hardware and software solutions to different scenarios. Finally, we exploited the side information obtained by communications systems for human sensing. This proposed system is part of a more general scenario of integrated communications and sensing that is gaining more relevance due to the features of next-generation communications networks. In the last chapter, we discuss this interesting double role of communication networks and we propose a framework to detect human positioning indoors with no hardware specific for sensing.