Page - 374 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

5. DataDistribution TheDataDistribution component role is to provide i) sets of labeled signals according to specific needs; ii) trained classifiers; and iii) labels corresponding to the activities performedgiven framesof signals. Theavailabilityofsetsof labeledsignalshelpsresearchers thatneedtovalidate their techniqueswithawideandpublicdataset.Thehugeamountofdataalsoallows towork with deep learning techniques, and publicly available datasets allows to generalize the results and tocompare techniques. Figure4sketches theDataDistributioncomponent architecturalorganization. Data Distribution DATASET X,Y,Z SAMPLE RUN LABEL UniMiB Tracker COMPONENT UniMiB Acquisition Deploys Push H T T P Classifier Builder H T T P Classifier Deployer Online Classifier H T T P Control and Data Flow CLP Formatted Dataset with Harmonized signals and Labels Module Trained Classifier Figure4. TheDataDistributioncomponent. TheClassifierBuildermodule is in chargeofdistributingactivity recognitionmod- els that can be integrated in domain dependant applications. Themodule also relies on theClassifier Deployer to store the new trained activity recognitionmodel to be used for online classification. Finally, theOnlineClassifiermodule provides online services related to classification: givena set of inertial signals, it provides information regarding the activity the subject is performing. Finally, a user asks for datasets by performing a HTTPrequest. Froman implementation point of view, theDataDistribution component is aweb serviceexposing thegetDataset,getModule, andgetLabel functions. 5.1. PreliminaryValidationof theDataDistributionComponent Thecomponenthasbeendesigned,but its implementationhasbeenpostponed in favour of the other components that aremore challenging. Till nowwehave implemented the request for setsof signals asaRestFul endpoint. 6. Conclusions Humanactivity recognition is a very challenging and active researchfield andhas seen a rapidgrowth in thepast fewyears. A.Ferrari etal. /AFramework forLong-TermDataCollection374