MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202621072786 A) filed by Prof. Mahesh Suryakant Bhosale; Mr. Yogesh Mayappa Gavade; Mr. Shivam Mahadev Dhanlobhe; Mr. Sarang Dhananjay Deshpande; and Mr. Vaibhav Sanjay Dhamane on June 12, 2026, for Road Sign Recognition System Using Cnn And Reinforcement Learning Guided Gan-Based Reconstruction For Damaged Road Signs.
Inventors include Prof. Mahesh Suryakant Bhosale; Mr. Yogesh Mayappa Gavade; Mr. Shivam Mahadev Dhanlobhe; Mr. Sarang Dhananjay Deshpande; and Mr. Vaibhav Sanjay Dhamane.
The application for the patent was published on June 26, 2026, under issue no. 26/2026.
Abstract: The present invention relates to a Road Sign Recognition System Using Convolutional Neural Network (CNN) and Reinforcement Learning-Guided Generative Adversarial Network (RL-GAN)-Based Reconstruction for damaged road signs. The invention is configured to detect, analyze, reconstruct, classify, verify, and generate alerts corresponding to road traffic signs under real-time vehicular and environmental conditions. The proposed system comprises an image acquisition module for capturing road scene images, a pre-processing module for image enhancement and road sign extraction, a damage severity estimation module for evaluating structural degradation, and a CNN-based classification module for traffic sign recognition and confidence score generation. An adaptive threshold evaluation mechanism dynamically determines classification acceptance criteria based on environmental conditions, image quality, visibility levels, and traffic sign degradation severity. When a traffic sign image produces a low-confidence classification or exhibits substantial structural degradation, the image is automatically redirected to an RL-GAN reconstruction module comprising a generator network, a discriminator network, and a reinforcement learning agent. The RL-guided reconstruction mechanism restores damaged, faded, blurred, incomplete, or partially occluded traffic sign regions while optimizing reconstruction quality through reward-based learning. A reconstruction verification module validates reconstructed outputs by comparing original and reconstructed confidence scores and evaluating semantic consistency prior to final classification. The reconstructed traffic sign image is subsequently reclassified and corresponding driver alerts are generated. The invention provides an intelligent, adaptive, and self-correcting framework that improves recognition accuracy, reconstruction reliability, environmental adaptability, and road safety, making it suitable for Intelligent Transportation Systems (ITS), Advanced Driver Assistance Systems (ADAS), autonomous vehicles, vehicular safety systems, and smart traffic monitoring applications.
Disclaimer: Curated by HT Syndication.
