International Journal of Science, Engineering and TechnologyAuthors: Dhamdhere Shubhangi Tulshidas, Sivaram Ponnusamy
Abstract: Personal safety technologies for women in public and semi-public spaces remain dominated by manually triggered panic-button applications, a design that depends on the wearer being conscious, unrestrained, and able to deliberately activate a device at precisely the moments when that is least likely. This paper reviews two artificial-intelligence-based wearable sensing approaches capable of detecting an emergency automatically, without a manual trigger: motion-based anomaly and fall detection, which recognizes sudden-impact or struggle-like movement using inertial sensors and deep human-activity-recognition models, and physiological stress sensing, which detects the autonomic-arousal signature of fear using heart-rate and skin-conductance signals already available on consumer wearables. The review identifies three related gaps: deployed safety products remain almost entirely manually triggered despite the maturity of automatic anomaly-sensing methods; motion-based and physiological-based detection have developed as separate research communities and are rarely fused into one behavior-risk score; and the always-on, battery- and privacy-constrained deployment context of a wearable is rarely addressed jointly with detection accuracy. To address these gaps, the paper proposes an integrated four-stage architecture, data acquisition, behavior and anomaly modeling, risk fusion and tier classification, and graduated emergency response, together with an illustrative four-tier behavior-risk scheme that replaces a binary silent-or-SOS design with escalating response actions. Following a literature-synthesis and conceptual-design methodology rather than a primary field trial involving human subjects, the results synthesize literature-reported performance for each modality and work through an illustrative, hypothetical wearable-deployment cohort to show how a graduated response would distribute across everyday use. The discussion highlights the unvalidated fusion gap and on-device privacy and battery constraints as the most consequential caveats, and the paper closes by outlining the consent-based validation studies and standardized, wearable-first benchmarks needed to move from the current, component-by-component evidence base toward a deployable, trustworthy safety wearable.
DOI: http://doi.org/10.5281/zenodo.23188805
