Seite - 367 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

AFramework forLong-TermData Collection toSupportAutomaticHuman ActivityRecognition AnnaFERRARIa,DanielaMICUCCI,a,1,MarcoMOBILIOa and PaoloNAPOLETANOa aUniversityofMilano -Bicocca Abstract. Human activity recognition (HAR) is a very active field of research andmany techniqueshavebeendeveloped in recent years.Deep learning techniques applied to3Dand4Dsignalssuchasimagesandvideoshaveproventobeeffective,achiev- ing significant classification accuracy. Recently, these techniques are also being usedfor1DsignalsandareexploitedtorecognizehumanActivitiesofDailyLiving (ADLs).Theyprocess inertial signals such as those obtained fromaccelerometers andgyroscopes.However, to compute accurate and reliabledeep learningmodels, a lot of sample data is required. Moreover, the creation of a dataset to be used with deep learning techniques is anonerous process that requires the involvement of a significant number of possibly heterogeneous subjects. The publicly avail- abledatasets are fewand,with rareexceptions, contain fewsubjects.Furthermore, datasets are heterogeneous and therefore not directly usable all together. Thegoal of ourwork is the definition of a platform to support long-termdata collection to be used in trainingHAR algorithms. The platform, termedContinuous Learning Platform(CLP),aimsto integratedatasetsof inertial signals inorder tomakeavail- able to thescientificcommunitya largedatasetofhomogeneoussignalsand,when possible, enrich itwithcontext information (e.g., characteristicsof the subject, de- viceposition, andsoon).Moreover, theplatformhasbeendesigned toprovidead- ditional services suchas thedeploymentof activity recognitionmodels andonline signal labelling services. The architecture has been defined and someof themain componentshavebeendeveloped inorder toverify the soundnessof theapproach. Keywords.dataset,deeplearning, inertialdata, integrationplatform,ADL,activity, machine learning 1. Introduction HumanActivityRecognition (HAR) is an active research field aimed at experimenting with newmethods and techniques for the automatic recognition ofActivities of Daily Living(ADLs)and, insomecases,alsooffalls[27,23,21].Mostoftheproposedmethods andtechniquesexploit sensorsembeddedinsmartphones,smartwatches,fitness trackers, andad-hocwearabledevices.Theclassificationofsensordatawithrespect to theactions 1CorrespondingAuthor:DanielaMicucci,University ofMilano -Bicocca,Viale Sarca 336, 20126Milan, Italy;E-mail: daniela.micucci@unimib.it. Intelligent Environments 2019 A. Muñoz et al. (Eds.) © 2019 The authors and IOS Press. This article is published online with Open Access by IOS Press and distributed under the terms of the Creative Commons Attribution Non-Commercial License 4.0 (CC BY-NC 4.0). doi:10.3233/AISE190067 367