MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202641110835 A) filed by Pranesh on September 16, 2026, for A Privacy-Preserving Edge-Deployable Gait Recognition System Using Federated Transfer Learning And Multi-Backbone Feature Fusion.
Inventors include Pranesh Vijayakumar; and N. Sivakumar.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: The invention presents a gait recognition system intended for distributed surveillance and edge-based use. A walking sequence captured at a local camera node is processed at the same location, so the original gait recording does not need to be transferred to a central server. The prepared gait sequence is analysed using several pre-trained deep networks, including AlexNet, EfficientNet-B0, DenseNet-121, and Xception. Their outputs are converted into a common feature form and combined to obtain the gait representation used for identification. Local nodes train the recognition model with their own data and forward only the learned model updates to a federated server. The server combines these updates and produces a shared model, which can be returned to the participating nodes for further learning. For edge use, the trained model may be reduced through pruning, quantization or knowledge distillation. The resulting model can then perform gait-based human identification on resource-limited devices while the original biometric gait data remain at the local source.
Disclaimer: Curated by HT Syndication.
