In this paperwe present the development of an\r\ninteractive,contentawareandcosteffectivedigitalsignage\r\nsystem. Using a monocular camera installed within the\r\nframe of a digital signage display, we employ realtime\r\ncomputervisionalgorithmstoextracttemporal,spatialand\r\ndemographic features of the observers,which are further\r\nused for observerspecific broadcasting of digital signage\r\ncontent.ThenumberofobserversisobtainedbytheViola\r\nandJonesfacedetectionalgorithm,whilstfacialimagesare\r\nregistered using multiview Active Appearance Models.\r\nThedistanceoftheobserversfromthesystemisestimated\r\nfrom the interpupillary distance of registered faces.\r\nDemographic features, including gender and age group,\r\naredeterminedusingSVMclassifierstoachieveindividual\r\nobserverspecific selection and adaption of the digital\r\nsignage broadcasting content. The developed system was\r\nevaluatedatthelaboratorystudylevelandinafieldstudy\r\nperformed for audience measurement research.\r\nComparison of our monocular localization module with\r\nthe Kinect stereosystem reveals a comparable level of\r\naccuracy. The facial characterizationmodule is evaluated\r\non the FERET database with 95% accuracy for gender\r\nclassification and 92% for age group. Finally, the field\r\nstudy demonstrates the applicability of the developed\r\nsysteminreallifeenvironments.
Loading....