MUMBAI, India, July 24 -- Intellectual Property India has published a patent application (202611063811 A) filed by Delhi Technological University on May 20, 2026, for System And Method For Detecting And Locating An Acoustic Event.
Inventors include Sood, Akshat; Patidar, Sanjay; and Sharma, Sumit.
The application for the patent was published on July 17, 2026, under issue no. 29/2026.
Abstract: A distributed gunshot detection and localization system (100) comprising a plurality of sensor nodes (110), each node including an infrared muzzle-flash detector (114), microphone (111) sampling at 48 kHz, and edge processor (112) executing a student convolutional neural network trained via teacher-student knowledge distillation. Each node transmits privacy-preserving event messages containing acoustic feature vectors and timestamps without raw audio data, having a size of less than 500 bytes per event, achieving a bandwidth reduction of at least 99% over raw audio streaming. A server (120) applies multi-node bimodal consensus requiring at least two nodes (110) to independently satisfy an if-and-only-if verification condition: both the acoustic confidence threshold and temporal coincidence between infrared and acoustic detections within 30 milliseconds must be satisfied simultaneously. This achieves a False Acceptance Rate (FAR) reduction of at least 80% compared to audio-only systems. The server (120) computes gunshot location using time-difference-of-arrival measurements corrected for per-node clock offset and drift rate, employing robust regression algorithms selected from Huber regression, Tukey’s biweight regression, an-md edleiaastn-of-squares regression that reject outlier measurements, achieving localization accuracy within ±5 meters in urban multipath environments. The system further supports federated learning of edge model parameters without transmission of raw audio data.
Disclaimer: Curated by HT Syndication.
