Tefnut: An Accurate Smartphone Based Rain Detection System in Vehicles

Real-time and fine-grained rain information is crucial not only for climate research, weather prediction, water resources management, agricultural production, urban planning and natural disasters monitoring, but also for applications in our daily lives. However, because of the lack of rain detection systems and the high variable attribute of rain, both in time and space, the rain detection today is still not precise enough. In such context, we propose and implement Tefnut (Tefnut is the rain deity in Ancient Egyptian religion.), a novel system that exploits opportunistically crowdsourced in-vehicle audio clips from an alternative, nowadays omnipresent source, smartphones, to achieve precise detection of rain leveraging a supervised recognizer constructed from a series of refined features. We conduct extensive experiments, and evaluation results demonstrate that Tefnut can detect the rain with 96.0% true positive rate, when deciding with a one-second-long in-vehicle audio segment only.