MUMBAI, India, Aug. 10 -- Intellectual Property India has published a patent application (202621072878 A) filed by Symbiosis International Deemed University on June 11, 2026, for Ai-Based Real-Time Traffic Violation Detection System Using Yolov8, Cnn, And Federated Learning.
Inventors include Madhura Hawelikar; Rohit Kindarle; Sushmit Partakke; and Dr. Gagandeep Kaur.
The application for the patent was published on July 31, 2026, under issue no. 31/2026.
Abstract: ABSTRACT AI-BASED REAL-TIME TRAFFIC VIOLATION DETECTION SYSTEM USING YOLOV8, CNN, AND FEDERATED LEARNING An AI-based real-time traffic violation detection system (100) integrating YOLOv8 object detection, CNN feature extraction, rule-based violation inference, and federated learning for automated two-wheeler traffic enforcement is disclosed. The system (100) comprises an edge perception layer (110) with IP dashcams (112) and edge computing devices (114), a YOLOv8n detection network (120) employing anchor-free detection with decoupled classification-regression heads for simultaneously detecting riders, helmeted riders, unhelmeted riders, and number plates, a violation inference engine (140) using spatial containment and IoU rules for helmet non-compliance, triple- riding, and licence plate absence determination, an evidence generation module (150) producing GPS-tagged timestamped evidentiary clips with OCR-derived plate records, a federated learning module (160) enabling privacy-preserving distributed model improvement via 16-bit quantised weight transmission, and a backend management layer (170) with enforcement dashboard. The system achieves 93.8 percent mAP@50 at approximately 25 FPS on edge hardware. [
Disclaimer: Curated by HT Syndication.
