MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202641101607 A) filed by Chennai Institute Of Technology on August 22, 2026, for Adaptive Privacy-Preserving Federated Fall Detection System With Clothing-Adaptive Pose Estimation.
Inventors include Dr. J. Brindha; Dr. R. Bino Wesley; Mrs. S. Sri Abirami; and Mrs. R. Ramya.
The application for the patent was published on August 28, 2026, under issue no. 35/2026.
Abstract: A federated learning system and method for privacy-preserving, vision-based patient fall detection dynamically selects, for each client device and communication round, among differential privacy, secure multi-party computation, and homomorphic encryption, based on a computed risk score derived from client computational capability, network conditions, data-sensitivity classification, and a client trust score, with a privacy budget scheduler allocating a differential-privacy budget non-uniformly across rounds and a tamper- evident, distributed-ledger-anchorable audit log recording the selection metadata. A sliding-window pre-processing module segments captured fall-image sequences into overlapping windows, and a clothing-adaptive pose-estimation module detects and corrects garment-induced occlusion of skeletal keypoints for subjects wearing sarees, dhotis, lungis, mundus, or shawls, prior to classification by a spatio-temporal graph convolutional network integrated with a two-phase federated feature-fusion and personalization framework (P2-FFI). Population-specific biomechanical fall-detection thresholds, derived from an Indian Elderly Activity Dataset (IEAD), are applied, and a multi-stage confirmation module corroborates a primary classification with a secondary, privacy-preserving signal before an alert is issued, reducing false positives. The complete pipeline is deployable, in quantized form, on resource-constrained edge hardware such as a Raspberry Pi 5 or NVIDIA Jetson Nano.
Disclaimer: Curated by HT Syndication.
