Hand-handled user devices, such as smartphones and tablets, as well as in-vehicle sensors are becoming extremely popular and represent a fertile ground for the development of crowdsensing (or, equivalently, community urban sensing) applications. Indeed, user devices are typically equipped with cameras, gyroscope, air-quality sensor, and accelerometer. The data obtained through such devices, as well as the ones gathered by in-vehicle sensors, can be combined with cellular BS positioning or GPS so as to obtain geo- and time-referenced samples of the urban environment. Cellular and WiFi interfaces can then be used, either interactively or autonomously, to transfer the user-generated data to the Internet, where they can be collected and processed. From such samples, it is therefore possible to estimate a large-scale phenomena of interest, such as air quality, noise level, as well as road traffic and vehicle speed.
Clearly, urban sensing applications pose several challenges, including user privacy guarantees, data credibility, incentives to users for employing crowdsensing applications and efficient use of the wireless resources for uploading data to the Internet. The latter two aspects in particular are related to the amount of information that should be collected and processed in order to estimate the phenomenon of interest accurately enough.
We focus on such an aspect and aim at identifying the number of samples that should be collected through handhelded devices as well as in-vehicle sensors, and transferred wirelessly to a processing center so that the phenomenon is reconstructed with an acceptable error. To do so, we take a signal processing approach and consider the problem of reconstructing the phenomenon (i.e., signal) from a set of samples that have been collected at irregular points in space, i.e., at the user/vehicle locations. Given the sample distribution, we apply a signal reconstruction method and evaluate the mean square error (MSE) between the original and the estimated phenomenon. We therefore investigate the dependency of the estimate accuracy on the number of samples, the phenomenon characteristics and the level of noise affecting the collected samples.
We evaluate our technique in an urban area, considering realistic city cartography and road traffic information.
Our results show that the accuracy of the measurements collected by the users is more critical than the sheer number of users. Also, we find that a small number of pedestrian users is sufficient to significantly improve the quality of the urban sensing provided by vehicles only.
Focus