Section: New Results

Middleware for Mobile Crowdsensing

Participants : Yifan Du, Valérie Issarny, Bruno Lefèvre, Françoise Sailhan.

Mobile Phone Sensing (MPS) offers a great opportunity toward the large scale monitoring of urban phenomena, such as the exposition of the population to environmental pollution. Indeed, mobile crowdsensing empowers ordinary citizens to contribute (whether pro-actively or passively) data sensed or generated from their mobile devices. It allows acquiring hyperlocal knowledge at scale, thanks to the proliferation of mobile devices and the ubiquity of wireless broadband connection. On-demand mobile crowdsensing is in particular a cost-effective service model for smart cities. Numerous sensor types embedded in today's smartphones contribute valuable quantitative observations about the urban environment (e.g., noise, temperature, atmospheric pressure, humidity, light, magnetism). The observations further come along with the related spatial and temporal data, which allows for the analysis of hyper-local environmental knowledge. However, mobile crowdsensing brings valuable knowledge only if a sufficiently large crowd contributes and if we overcome the relatively low accuracy of the gathered data. This is the focus of our research.

We have in particular studied how to reduce the gap between the need for the massive collection of relevant data, and the quantity and accuracy of the measurements that are actually gathered. We specifically carried out an iterative research process to tackle this challenge, which combines technological innovation and social design. We have been developing a number of social tools to study the motivations and usages of MPS-based smart city apps, with the Ambiciti app serving as our use case. Our study has been taking into account the cultural and societal contexts that the usages of Ambiciti could feed, spanning health, environment, education, and urban policies. We carried out an online survey together with interviews with users and local actors in Europe, i.e., France, Belgium, and Finland. The research results contribute to a better understanding of why and how people use mobile phone sensing applications; the results also inform how to best leverage mobile crowd-sensing in the development of smart cities and how it may serve addressing urban challenges related to, e.g., public health or urban planning.

The quality of the contributed measurements challenges the aggregation of relevant knowledge from crowdsensed observations. The measurements quality depends on the accuracy of the contributing sensors and the adequacy of the sensing context. Addressing the former relies on the sensor calibration for which we study both micro- and macro-level solutions. Addressing the latter requires a supporting inference mechanism, for which we introduce a personalized hierarchical inference of all the context elements that are relevant to the phenomenon that is monitored through crowdsensing, and under which the crowdsensor operates. This enables accounting for the specific behavior of the contributing end-user across time, as well as for all the features -and only those- that are relevant and locally available, while reducing the feedback required from the user for the personalization.