A wearable, ambient sound-based approach for infrastructureless fuzzy proximity estimation

While it is challenging to obtain absolute location information of people and objects in indoor venues, obtaining proximity information is sufficient in various wearable computing applications. In this work, we characterize to which extent sound - a modality available in any mobile devices - may be used to infer proximity between these devices. We introduce a fingerprinting based approach and verify the existence of a relation between the distance of two devices and the similarity of the recorded ambient sound. While a quantitative distance estimation could only be achieved with an accuracy of 46 %, we could infer a correct distance region with an accuracy of 80 %. We further showed that an accurate estimation which of two devices is closer to a reference device is possible and that we could reliably infer if two devices are at the same location.