With the immense popularity of smart phones, people now have devices that make large-scale sensing possible using internal sensors and the Internet connection. This allows to perform a large number of measurements with little cost, covering a large area and a long time span, because the sensor hardware is provided by the participants. This opens possibilities for use cases where traditional sensor networks are not applicable.
While dedicated sensor nodes can have a certain accuracy, participants of crowd sensing are unreliable or even malicious. Consequently, individual measurements are unreliable. But the combination of many measurements makes it possible to identify false data and increase accuracy.
We investigate this issue using the example measuring noise pollution caused by trains or planes. These typically produce specific noise patterns, so that using multiple sensor nodes these can be differentiated from other noise. Typically, smart phones can measure the noise level using the internal microphone, and determine their precise location and the exact time using GPS. There have been several approaches to measure noise pollution using crowdsensing, which have shown the practical applicability of this method for this use-case. But the measurements depend on the reliability and accuracy of every participating node.
By means of simulations, we analyze how the quality of information can be increased by combining the measurements of different participants or introducing reliable and accurate sensor nodes, in order to reduce the measurement error and identify unreliable nodes.