MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641119338 A) filed by G. Hemaprabha; Vishak A; Srinithi K; Vishal Dharsan Sr; Visha I Raj N D; and Yashvanth K on October 06, 2026, for A Multi-Modal Artificial Intelligence Monitoring And Companion System For Dementia Patients.
Inventors include G. Hemaprabha; Vishak A; Srinithi K; Vishal Dharsan Sr; Vishai Raj N D; and Yashvanth K.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: he present invention pertains to a computer-based multimodal artificial intelligence monitoring and companionship platform, DementiaCare OS, for dementia care environments. The platform employs a client-server architecture in which a client device captures patient video and audio and transmits the signals to a centralized server for processing. The server hosts a computer-vision module for fall detection, wandering detection, and face recognition using pose-estimation, trajectory- tracking, and face-recognition models, respectively, each generating structured events containing confidence scores and bounding-box metadata. In parallel, a speech- processing module performs automatic speech recognition (ASR) transcription and acoustic distress classification to identify distress indicators and repetitive- questioning patterns. The platform further includes an authentication and access-control subsystem that authenticates users via credential-based verification and issues a signed authorization token, with the caregiver-facing dashboard enforcing role-based access control distinguishing primary caregiver and read-only observer roles. Each structured event is persisted to a dedicated event store for retrospective behavioural analysis and audit review, while events whose confidence score falls below a predetermined threshold are routed to a caregiver-review queue rather than generating automated alerts, enabling human-in-the-loop adjudication. The platform further includes threshold-based detection of sundowning-related behavioural changes, a conversational companionship module for reminiscence therapy and repeated-question handling, and a fallback mechanism routing natural-language requests to a cloud-based large language model when local computational resources are insufficient. By integrating multi-modal sensing, schema-driven event fusion, confidence-gated human review, secure role-based access, safety-guardrailed reasoning, and persistent event storage within a unified thin-client architecture, the invention facilitates practical, scalable, and non-invasive monitoring and companionship support for individuals with dementia.
Disclaimer: Curated by HT Syndication.
