Evaluating Activity Recognition Diversity as Readiness Measure Towards Mainstream Applicability

Abstract

The demand for autonomous activity recognition (AR) has intensified courtesy of expanding avenue of uses especially in social media services. As existing ARs are designed for static activity instances recognition needs; the inherent lack of activity diversity has inevitably limits their prospect for mainstream applicability. The objective of this survey paper is to consolidate that activity diversity is a limitation holding activity recognition back and evaluates existing ARs’ readiness towards open world implementation. For this purpose, we conducted survey and formulate the perception values of existing methods in layman-centric environment over multiple location settings. We do this by introducing a diversity index metric to gauge the effectiveness of existing ARs in academic, sports, home and office environments given their current activity recognition depths. Our survey outcome discovered that existing ARs employed static and small activities set for curated purposes, which has a small hit ratio compared to real world ADLs. Our findings shows that existing ARs technology despite their acclaimed robustness of, they are only effective in respective curated domains and needs to evolve in terms of activity diversity support for universal appeal towards mainstream uses. The paper concludes with discussion on potential uses of IPv6 technology as the catalyst to account for diversity towards public friendly and ready ARs technology going forward.