Visual pollution localization through crowdsourcing and visual similarity clustering

Nowadays, many cities and communes suffer from advertisements appearing on aesthetically inappropriate or illegal places. This contamination of public space is called visual pollution. The first step in the fight against visual pollution is localization of physical advertising media (e.g., billboards) as accurately as is possible. One of the ways is to use volunteer effort through outdoor crowdsourcing. Smart mobile devices can support this process through localization sensors. However, these sensors are inaccurate enough on their own, plus, the media are not located exactly where the volunteers capture them. Therefore, the media localization is presently inaccurate. This paper presents a work-in-progress method to improve the localization of physical advertisement media. As input, the method takes captured media images along with spatial information about the device. The images are then clustered based on their locations, to form sets corresponding to the true physical media. Then, using visual analysis of the images and spatial orientation of devices, the method computes expected location of the physical media.